Hudson River PCBs Site
EPA Phase 1 Evaluation Report
Addendum
Prepared for:
US Environmental Protection Agency, Region 2
s^tD Sr%
Q
PRO"^
and
US Army Corps of Engineers, Kansas City District
Prepared by:
The Louis Berger Group, Inc.
April 2010
-------
Table of Contents
INTRODUCTION
TOPIC 1: WATER COLUMN LOADS DURING AND AFTER DREDGING
1 -A PCB CONCENTRATIONS AT THE FAR-FIELD STATIONS: POST-
DREDGING VS BASELINE
1 -B NEAR-FIELD PCB RELEASE MECHANISM STUDY
1-C NEAR-FIELD PCB TRANSECT STUDY
1 -D PCB FATE AND TRANSPORT IN THE FAR-FIELD DURING DREDGING
TOPIC 2: CAUSES OF RESUSPENSION DURING PHASE 1 DREDGING
TOPIC 3: REDISTRIBUTION OF CONTAMINANTS DURING DREDGING
TOPIC 4: FORECASTS OF DREDGING RELATED IMPACTS AND RELATED
RISK
4A FORECAST OF PCB LOADS DUE TO DREDGING COMPARED TO
NATURAL ATTENUATION
4-B PREDICTION OF IMPACTS ON FISH TISSUE CONCENTRATIONS IN
THE LOWER HUDSON RIVER DUE TO DREDGING
TOPIC 5: SCOW UNAVAILABILITY AND ITS IMPACT ON PRODUCTIVITY
TOPIC 6: UNDERESTIMATION OF DEPTH OF CONTAMINATION AND ITS
IMPACTS ON THE PROJECT
6-A POST-DREDGING CORE LOCATION TREATMENTS AND THE NEED FOR
MULTIPLE RESIDUAL PASSES
6-B RELEVANCE AND CONSEQUENCES OF UNCERTAINTY IN
MEASUREMENTS OF THE DEPTH OF CONTAMINATION
TOPIC 2 ATTACHMENT: CONDITIONS ASSOCIATED WITH WATER COLUMN
PCB CONCENTRATIONS: THOMPSON ISLAND DAM
2009
Hudson River PCBs Site Page i
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
INTRODUCTION
-------
INTRODUCTION
This Addendum to the EPA Phase 1 Evaluation Report presents information compiled in
response to requests and questions submitted by the Peer Review Panel after its introductory
meeting in February 2010. This Addendum also includes several evaluations referenced in
EPA's Phase 1 Evaluation Report but not yet completed at the time of its release. The contents
of this Addendum corroborate and supplement the conclusions presented in the report, support
EPA's proposed changes to the Engineering Performance Standards (with certain adjustments
which are discussed below or which will be explained at the Peer Review Deliberation Meeting),
and help demonstrate that the dredging operation can be modified in a number of respects to
obtain significant improvements over the Phase 1 performance.
The Addendum is a series of independent discussions organized under six Topics to provide a
logical sequence of information. The findings are summarized below:
1. Water Column PCB Loads During and After Dredging - The release of PCB non-aqueous
phase liquid (NAPL) from the dredging areas is influencing the resuspension monitoring
results; however, it is difficult to accurately quantify its impact. The presence of NAPL is
consistent with the high dissolved PCB concentrations detected in combination with low TSS
concentrations. Changes observed in homolog patterns between near and far-field samples
can be attributed to gas exchange of lighter, more volatile PCBs with the atmosphere during
transport downstream. TSS data do not support attribution of resuspension monitoring
results to releases and subsequent transport of suspended sediments. GE's resuspension data
interpretation and analysis do not incorporate these findings.
2. Causes of Resuspension - The most significant factors affecting resuspension are not simply
the mass removed and river velocity. A group of six factors comprises the key "resuspension
drivers": mass and volume removal, vessel traffic, exposure of recently disturbed dredging
residuals to flow, backfilling processes, and bucket closure.
3. Redistribution of Contaminants - Resuspension impacts are not exclusively tied to the
creation and transport of suspended solids releases from the dredging area. As discussed in
Topic 1, TSS data do not support the conclusion that significant amounts of sediment were
redistributed beyond the dredging areas. Sediment trap data analyses advanced in the GE
Phase 1 report lack a robust baseline data set, and were also impacted by tug traffic; they are
therefore not reliable and the co-located cores used by GE as surrogates for this analysis are
not adequate to support their assertions.
4. Forecasts of Dredging-Related Impacts and Related Risk - EPA has revised the analysis
set forth in Appendix I-G of EPA's Phase 1 Evaluation Report to improve the predictions of
PCB loads at Waterford. The reanalysis used PCB concentration instead of loads, since a
better correlation could be developed for concentrations to evaluate the recovery. The
analysis indicates that a dredging release of approximately 500 kg of Tri+ PCB, as measured
Hudson River PCBs Site Page 1
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
at the Waterford far-field station, provides net benefit to the river compared to MNA. While
the load standard will continue to be based on a 1 percent net loss of PCB mass removed
(approximately 2000 kg Total PCBs or 667 kg Tri + PCBs), the goal will be to minimize the
dredging-related release of Tri+ PCB to 500 kg (as measured at Waterford) over the life of
the project (including both Phase 1 and Phase 2). Support for the adoption of control levels
derived from the 1 percent net loss figure is provided by consideration of the impact on fish
tissue concentrations and project risks in both the Upper and Lower Hudson River. Analysis
was performed to estimate the impact of dredging-related PCB release on fish tissue
concentrations in the Lower Hudson River for the expected duration of dredging and the brief
period immediately beyond. The Farley-FISHRAND models were run for Upper Hudson
loads that correspond to the upper end of the proposed revisions to the Resuspension
Standard. The impacts predicted here are similar to those noted in the supporting
documentation for the Record of Decision. This analysis indicates that the proposed range of
revisions to the allowable load standard for the Lower Hudson would not have any
measurable long term impacts to the recovery of fish tissue concentrations and therefore
would not pose any additional long term risks to Hudson River fish-consuming populations
in the Lower Hudson.
5. Scow Unavailability and Its Impact on Meeting Performance Standards - Dredging
project implementation can and should be enhanced during Phase 2. The lack of availability
of empty scows (due to only partial filling and limited unloading capacity) created a greater
adverse impact on dredging productivity than did resuspension concerns. It also increased
boat traffic and the length of time CUs were left open, both of which contribute to
resuspension. The addition of a second unloading station and the addition of a minimum of
another 0.5 foot of dredged material to each scow will be necessary to facilitate achievement
of the Phase 2 dredging productivity targets.
6. Underestimated Depth of Contamination: the Need for Multiple Dredging Passes - EPA
shows that GE's use of SSAP cores is incorrect and causes an underestimate of PCB mass
removed during Phase 1 dredging. Most of the additional dredging was required to address
underestimation of the depth of contamination (DoC). Since the additional dredging was
generally not conducted to capture dredging residuals but rather inventory, the addition of a
designed overcut to the first dredging pass during Phase 2 will reduce the number of
dredging passes required and increase efficiency. An overcut is needed to address
uncertainty in coring data and the dynamic nature of the sediment bed.
These findings corroborate EPA's belief that Phase 2 of the dredging remediation can be
accomplished in a way that achieves the benefits ascribed in the ROD, with some modifications
to the EPS and the dredging project design, and that it can consistently meet EPA's proposed
engineering performance standards for Phase 2.
Hudson River PCBs Site Page 2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Below is a summary of each of the Topics presented in the Addendum. Also discussed below is
a recent decision by EPA relating to the Halfmoon and Waterford public water supplies, and the
implications of that decision for the use of the 500 ng/L water column standard.
The low detected TSS concentrations, combined
with high dissolved concentrations in the near-
field studies, indicate that release of PCB NAPL
is influencing the resuspension monitoring
results (although the impact cannot be accurately
quantified). EPA's evaluations of GE's water
column studies found that near-field releases
were dominated by PCBs in the dissolved phase,
and that the data do not support GE's conclusions that the oil phase observed near the dredging
operations represented only small, ephemeral amounts of PCBs. A 2-phase partitioning model as
used by GE is not applicable in the presence of NAPL and will incorrectly identify the PCB
partitioning as being in equilibrium. Since the NAPL was not explicitly characterized and is not
appropriately represented in GE's model, its overall impact on the dissolved phase
concentrations cannot be quantified. If the NAPL is as concentrated as oil found at the Fort
Edward Plant Site former 004 Outfall #5-58-004 (GE, 2008), even a small droplet can
significantly alter the PCB concentration of a water sample. At the Fort Edward outfall, PCB
concentrations of 255 to 1,240 ppm (more than six orders of magnitude - a million times -
greater than concentrations in typical baseline river water samples) were detected in oil that is
similar to the oil which is the source of the NAPL in the river.
The dynamics of TSS and PCB loads in the near-field are highly variable due to the impacts of
dredging and related activities being conducted in the river. For this reason, samples collected
by GE along near-field transects can only be used to represent conditions during a relatively
short duration immediately preceding sample collection. It follows that load estimations based
on transect samples represent only conditions that occurred during a narrow time interval {i.e., a
snapshot), and cannot be used to create a long-term mass balance to draw conclusions about
transport of PCBs from dredging. The variability of the near-field was recognized when the
Resuspension Standard was developed and, at that time, it was determined that the only
substantive, reliable measure of load is data from far-field stations. GE's observations from
Phase 1 confirm that the near-field is highly variable.
The PCBs detected in the far-field samples at Thompson Island contained a greater percentage of
Tri+ homologs (45 percent) than the near-field PCB transect study samples (31 percent) and the
near-field release mechanism study samples (18 percent). It is most likely that, during transport
from the near-field, the lighter, more-volatile PCB homologs were lost by gas exchange, causing
the shift towards higher Tri+ PCB fraction at Thompson Island. Loss of PCBs to the air was
documented on a number of occasions during Phase 1, particularly when an oil phase was visible
Topic 1 is an examination of the water
column loads observed during and
after the 2009 Phase 1 dredging and an
analysis of various studies regarding
water column PCB and TSS loads that
were conducted by GE.
Hudson River PCBs Site Page 3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
and concentrated, such as at CU-3 or in the sheet pile area at CU-18. While gas exchange losses
were most apparent in these focused areas, the elevated local concentrations near the dredge
operations likely create similar opportunities for gas exchange. EPA's hypothesis and analysis
showing a loss of the lighter homologs is consistent with the entire set of available data. GE's
hypothesis that a gain in Tri+ PCBs between the near-field and the far-field TID station is due to
transport of resuspended sediments is not supported by the observed TSS concentrations in the
near-field. TSS concentrations detected at monitoring buoys downstream of the East Griffin
Island Area (EGIA) were higher and more variable than TSS concentrations at Thompson Island.
If additional solids were re-suspended downstream of EGIA, the observed TSS concentrations at
Thompson Island would have been significantly higher than values at EGIA.
Five years of baseline monitoring established a fairly predictable relationship among PCB loads
at the far-field monitoring stations, with both TSS and PCB loads increasing steadily
downstream. During dredging, however, a consistent decline in all PCB homolog loads between
Thompson Island and Waterford was observed, despite an increasing TSS load downstream,
consistent in direction and magnitude with the observations of TSS under baseline.
EPA analyzed the loss of PCBs between Thompson Island and Schuylerville (Lock 5) during
dredging. EPA's findings suggest that recycling of particles at the sediment bed is resulting in
dilution of the Tri+ PCB concentration and a decline in Tri+ load, as well as the loss of lighter
fractions via gas exchange. It should be noted that the Total PCB/Tri+PCB ratio for the three far-
field stations consistently decreases with increasing flow. Because sediment scour and
resuspension are more likely under higher flow conditions, particle recycling with less
contaminated sediments in the far-field would maintain the high Tri+ component of TPCB, while
loss of monochlorobiphenyl and dichlorobiphenyl would ensure that the ratio increases
downstream. GE's hypothesis that the loss of PCBs is due to setting of particulate matter is not
consistent with the increases in the observed TSS loads between the two stations. While none
of these hypotheses can be unequivocally confirmed from the available data, any settling of
contaminated particles during transport from Thompson Island will not significantly change the
downstream contaminant exposure in surface sediment, as these materials will primarily settle in
the areas slated for dredging, since they are the primary areas for deposition and hence have
identified for removal..
Comparison of post-dredging water quality PCB data to baseline concentrations is hampered by
the fact that the TI station auto-sampler was significantly fouled sometime after November 2009
and at least 4 of the 5 sample ports were compromised under a load of debris and vegetation.1
All of the data from this sampler after November 2009 are suspect due to this condition. The
other auto-sampler stations have consistent Total PCB to Tri+ PCB ratios through time
1 The last time the ports were inspected prior to retrieval in mid-April 2010 was November 2009, and it cannot be
known precisely when the fouling occurred after that time. Freeze-thaw cycles which release vegetation to the river,
or accumulation of debris during high flows in January or March are possible explanations for the fouling. It is
believed that this fouling was sufficient to pull these intakes to the river bottom, especially under high flows.
Hudson River PCBs Site Page 4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
regardless of absolute concentrations; but at TID, the ratios (not just the concentrations) fluctuate
significantly over short periods of time, indicating there were problems with the sampler. At
Waterford, with the exception of one questionable data point, there are no indications of post-
dredging impacts after November 2009. At Schuylerville, there are possible indications of post-
dredging impacts during high flow conditions; however, data from some samples are suspect and
three of the five auto-sampler intakes at Schuylerville were compromised as well. . It appears
that GE's assertion of a significant post-dredging increase is tied to use of a "correction factor"
to the post-2009 analytical data, without applying the same correction factor to the baseline data.
This one-sided correction increases the post-dredging data without a corresponding increase in
the baseline, hence artificially indicating a post-dredging impact. Moreover, EPA has not
approved this correction factor because there is insufficient explanation for its basis and
differences from earlier evaluations.
EPA's careful analysis of factors associated with
water column PCB concentrations has found that
the primary "resuspension drivers" are associated
with 6 process variables. The most likely factors
contributing PCBs to the water column are not
unexpected—mass and volume removal, vessel traffic, exposure of freshly disturbed residual
sediments to active flows, processes associated with backfilling, and bucket closure. A multiple
variable factor analysis and corresponding regression has led to development of a factor-based
predictive model and shows that it is the combination of these variables that causes the
resuspension to be increased. By controlling process variables such as vessel traffic, bucket
closure, and the time CUs are open, Phase 2 resuspension and productivity will be improved
relative to 2009 performance. This analysis also indicates that Phase 2 dredging can be
conducted with reasonable resuspension rates while meeting productivity goals.
Statements in GE's Phase 1 Evaluation Report that exclusively predict resuspension in direct
proportion to PCB mass removed and flow are not a reasonable response to the extensive
information that is available to further refine and optimize the dredging operation. Recent
statistical analysis by EPA has shown that this view is over-simplified; the many and varied
mechanisms associated with increased water column PCB concentrations should not be reduced
to a mere proportionality to mass or volume removed. Mass or volume removed is a surrogate
for the net effect of all of the processes involved in dredging, and therefore correlates well with
water column PCB concentrations; however, the manner in which dredging and other related
activities were conducted during Phase 1 is very important and its examination provides a good
source of information on how dredging and other activities could be modified to reduce
resuspension of PCBs in Phase 2.
Topic 2 is an evaluation of the
factors that contributed to
resuspension.
Hudson River PCBs Site Page 5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
EPA's examination of the impacts of re-distributing
sediment downstream is in disagreement with GE's
argument that resuspension is evidence of the
distribution of large amounts of residuals to the sediment
bed beyond the dredge prisms. GE claims that, in the
near-field, 35 percent of the dredged PCB mass is lost. Although the EPA does not dispute that
some amount of resuspended sediment will be deposited outside of the dredge prism, this
assertion is difficult to explain in light of GE's denial of the presence of a significant oil fraction,
and the fact that there is no associated TSS increase. Part of GE's argument relies on the use of
sediment traps near the dredging during Phase 1. Apparently GE used sediment traps to collect
baseline sediments before the start of dredging but no analytical results are available. The lack
of baseline sediment trap data seriously affects the conclusion that can be drawn from the study
because there is no comparable baseline for sediment trap data. To compensate for the lack of a
baseline for the sediment trap data, GE collected post-dredging cores co-located with pre-
dredging SSAP cores for comparison to serve as surrogates. Comparison of pre-dredging surface
concentrations to post-dredging concentrations in the co-located cores showed no statistical
difference from comparisons among co-located core pairs where both cores were collected
before dredging started, as discussed in EPA's Phase 1 Evaluation Report. Thus, the co-located
cores simply cannot be used for this purpose. EPA's evaluation found that the sediment trap
deployment was significantly affected by tug traffic, and therefore is not reliable, as it is not
representative of general conditions.
The goal of the remedy as stated in the Record of
Decision (ROD, p. 50) is to "reduce the cancer risks
and non-cancer health hazards for people eating fish
from the Hudson River by reducing the
concentration of PCBs in fish." This goal is focused
on the improvement of conditions in the Upper
Hudson since this is where the greatest risks were
estimated and the greatest concentrations of PCB
are found. It was anticipated by the ROD that
remediation conducted in the Upper Hudson would also yield improvements in the Lower
Hudson but this improvement was not the primary goal of the remediation. EPA evaluated the
potential effects of the loads anticipated for Phase 2 by estimating the impacts to fish tissue
concentrations in the Lower Hudson.
The analysis presented in Topic 4-A of this Addendum revises the analysis set forth in Appendix
I-G of EPA's Phase 1 Evaluation Report, which stated that estimates of Total PCB loads at
Waterford based on USGS and GE data for the period 1995 to 2008 show no statistically
significant decline, with a "half life" of 99 years. The reanalysis uses PCB concentration instead
of loads to improve the original calculation, since a better correlation could be developed for
Topic 3 is an examination of
the re-distribution of PCBs as a
result of dredging.
Topic 4 examines the potential
impacts on fish tissue
concentrations in the Lower Hudson
River as a result of dredging as a
measure of the general impacts of
the remedy down-river.
Hudson River PCBs Site Page 6
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
concentrations to evaluate the recovery. Flows were simulated via a Monte Carlo analysis rather
than forecast as a single set of values. Additionally, further analysis of the historical data
suggested that the records for the years 1999 and 2000 were inconsistent with the general trend
observed in the data. Although no specific reason could be identified for the inconsistency, these
data were considered unusable and were excluded from the best estimate calculation of the rate
of concentration decline with time.
Predictions were made of Monitored Natural Attenuation (MNA) of Tri+ PCB load at Waterford,
generated from historical data from the United States United States Geological Survey (USGS;
1995-2001) and baseline monitoring data collected by General Electric Company (GE; 2004-
2009) using an empirical approach. The results of this analysis indicate a dredging release
limited to approximately 500 kg of Tri+ PCB, as measured at the Waterford far-field station,
provides a net benefit to the river based on an overlap in the band of forecasted uncertainty
between this dredging release scenario and MNA. While the load standard will continue to be
based on a 1 percent net loss of PCB inventory removed (approximately 2000 kg Total PCB or
667 kg Tri + PCBs), the goal will be to minimize the dredging-related release of Tri+ PCB to
500 kg (as measured at Waterford) over the life of the project (including both Phase 1 and Phase
2). Support for the adoption of control levels derived from the 1 percent net loss figure is
provided by consideration of the impact on fish tissue concentrations and project risks in both the
Upper and Lower Hudson River.
An analysis was performed to estimate the impact of dredging-related PCB release on fish tissue
concentrations in the Lower Hudson River for the expected duration of dredging (2009 and 2011
to 2015) and the brief {i.e., 5-8 year) period immediately beyond. The intention was to examine
potential incremental increases in projected fish tissue PCB concentrations. Additional risks
(above current risk levels) to human health via fish consumption would be identified if such
incremental increases in fish tissue concentrations remain elevated above the forecasted MNA
recovery curve for the remedy selected in the ROD for many years after completion of the
dredging. The water column and sediment concentrations in the Lower Hudson River were
estimated from the same model of chemical fate and transport that was used in the Performance
Standard (Farley et al. (1999)), referred to herein as the Farley model. The Farley model was run
for two dredging-related PCB release scenarios that correspond to the upper end of the proposed
revisions to the Resuspension Standard, to estimate sediment and water column concentrations of
the Lower Hudson out to 2046. The fish tissue concentration was predicted by interpolating
previous results from a bioaccumulation model developed for the Hudson River remedial
investigation and used in both the ROD and the Performance Standards (FISHRAND, USEPA,
2000).
Scenarios were run for dredging-related loads of 600 kg and 800 kg of Tri+ PCBs
(approximately 1800 and 2400 kg Total PCBs) at Waterford. The impacts of these loads were
contrasted with the MNA recovery curve, used here as a surrogate for the remedy curve response
Hudson River PCBs Site Page 7
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
in the Lower Hudson. This approach provided an upper bound on the likely time to rejoin the
ideal recovery curve (representing no resuspension load due to dredging). The results show only
minimal increases in the time to rejoin the ideal recovery curve relative to scenarios representing
less PCB loss to the Lower Hudson. The potential impacts due to dredging resuspension were
recognized as part of the preparation of the ROD and were included in the Responsiveness
Summary. The impacts predicted here are similar to those noted in the supporting ROD
documentation. Taken together, the observations and calculations from this analysis indicate that
the proposed range of revisions to the allowable load standard for the Lower Hudson would not
have any measurable long term impacts to the recovery of fish tissue concentrations and
therefore would not pose any additional long term risks to Hudson River fish-consuming
populations in the Lower Hudson.
The ability to meet productivity targets was
severely limited during Phase 1 by the
unavailability of scows. The reason for the lack
of scows is not due to the number of scows that
were on hand, but rather the way the scows were
used and how they were unloaded, which resulted
in long queues of partially full scows waiting to
be unloaded. Although there is significant discussion on GE's part about how resuspension
limited productivity, the reality is that the unloading system was under-designed for the Phase 1
sediment volume targets and scow unavailability is the most important factor that limited
productivity. Analyses by EPA show that productivity was decreasing due to scow queues
before resuspension-related shut-down events occurred (indicating other causes, as discussed
below) and that, given the optimum unloading rates recorded during Phase 1, the unloading
facility could not have sustained the target productivity.
EPA's evaluation shows that as GE attempted to meet the Phase 1 productivity target in late July
and early August 2009, productivity became limited by scow unavailability (as seen by the
number of scows in the unloading dock queue reaching a maximum) which caused the daily
volume dredged to start its decline two weeks before the early August resuspension event. This
contrasts sharply with GE's assertion that resuspension is the limiting factor for productivity. It
was found that in the days leading up to the shutdown as few as 4 scows were available for
loading by the 11 active dredges in 10 open CUs. Further, it was found that after the 4-day
Labor Day weekend, during which no dredging occurred, most of the scows had been unloaded
and were available to the dredgers {i.e., there were no scows waiting to be unloaded at the
unloading dock), and the two greatest productivity days of Phase 1 occurred. However, the
queue quickly returned to previous levels and productivity declined. In addition, scow queues
also increased the length of time CUs were left open, which contributed to resuspension, as did
GE's frequent practice of transporting under-filled scows (resulting in more vessel traffic than
would otherwise have occurred, as well as lengthening the queue).
Topic 5 examines the limitations to
productivity caused by system factors
such as scow load thickness and using
a single unloading station.
Hudson River PCBs Site Page 8
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
A probabilistic queuing model was developed using the Phase 1 data to evaluate productivity
based on load thickness and to make predictions of productivity that may be attained in Phase 2
using one and using two unloading stations. The model shows that, even with increased load
thickness, scow queues will limit productivity to less than the Phase 2 targets if there is only one
unloading station. However, by adding a second unloading station and loading scows with an
additional 0.5 foot of material (well within the load constraints of the scows) the productivity of
targets for Phase 2 can be met.
The uncertainty in DoC is acknowledged as one
of the largest impediments encountered during
Phase 1. GE asserts that much of the dredging
was spent attempting to remove residuals;
however, while additional dredging may have
been dictated by requirements included in the Residuals Standard, the fact is that the sampling
required by the standard identified additional inventory to be removed as a result of consistent
underestimation of the DoC. In the 10 CUs encompassing approximately 48 acres that were
dredged during Phase 1, 445 locations were targeted for post-dredging sampling. Notably, less
than 35 percent of the post-dredging coring sites had a residuals layer requiring removal. By
inference, less than 35 percent of the Phase 1 dredging area was re-dredged to address a residuals
layer. The purpose of this tracking and analysis performed by EPA was to measure the success of
the Standard in avoiding one or more residuals passes, and to demonstrate that, contrary to what
GE has asserted in its Phase 1 Evaluation Report (GE, 2010), over 60 percent of the remediation
area went directly from inventory removal to compliant, whereas only 8 percent of the Phase 1
area required 2 dredging passes. Moreover, since 6 inches of remaining inventory could not be
discerned from a true residual layer based on the Performance Standard sampling requirements,
the estimate of 35 percent is an upper bound on the actual extent of residuals. Ultimately this
analysis shows that the underestimated DoC was the primary cause of additional dredging
passes. Specifically, 42 percent of the 445 locations were shown to have inventory present after
the first dredging pass based on identification of deeper contaminated sediments during one of
the subsequent sampling surveys and 20 percent of all locations actually required three inventory
passes.
The tracking and analysis of each post-dredging coring location yielded several important results
pertaining to the implementation of the Residuals Performance Standard:
• It identified areas where the DoC was poorly characterized and inventory remained below the
first dredging pass (volume removed by inventory dredging was more than double of that
removed by residuals dredging).
• It provided evidence that the standard was successfully implemented, without undue
difficulty or schedule impacts, and that dredging-generated residuals (i.e., residuals due to
spillage and bucket movements) were not a pervasive issue.
Topic 6 evaluates the implications of
the underestimation of the depth of
contamination (DoC) during Phase 1.
Hudson River PCBs Site Page 9
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
• It provided further support of EPA's proposal for an overcut addition to each dredging pass
in Phase 2 to address uncertainty. Adding an overcut to the first pass will serve to reduce the
number of dredging passes and increase dredging efficiency, and will not simply produce
more residuals.
The findings presented in the Addendum corroborate EPA's belief that Phase 2 can be
accomplished in a way that achieves the benefits ascribed in the ROD, with some modifications
to the Engineering Performance Standards and with modifications to the dredging design. The
Addendum studies show that assertions of the inevitability of conflicts among the three
engineering standards are superficial and depend on over-generalization of the data. When the
details are more closely examined, the generalizations are not borne out. As an example,
contrary to GE's assertion that productivity in general is inextricably linked to resuspension,
there are several periods during Phase 1 where productivity was increased but resuspension was
not above critical thresholds. As shown in the Addendum, the severity of water quality impacts
due to resuspension (as measured at Thompson Island) is better explained by where dredging
was occurring, the volume of ship traffic, dredging bucket efficiencies, and river flow, than
simply by how much sediment (or PCB mass) was being dredged. The same dismantling of
generalizations occurs when examining assertions about residuals (e.g., distribution of residuals
outside of the dredge prisms, and the effects of uncertainty in depth of contamination),
resuspension (e.g., the use of near-field data to characterize loads attributable to dredging, and
attributing the TPCB to Tri+PCB ratio change with distance from dredging to settling alone), and
productivity (e.g., peak productivity was actually limited by scow unavailability due to
inefficiencies in unloading causing long scow queues, and not due to resuspension).
Water Supplies and the Resuspension Standard
EPA has informed the Towns of Halfmoon and Waterford, NY that it will agree to reimburse
them for their incremental costs in obtaining drinking water from the City of Troy until
November 2012, and throughout each of the Phase 2 dredging seasons, until the Hudson River
dredging is complete. This decision ensures that those towns, which are the only ones that have
a public water intake in the Upper Hudson, will not be using the river as their source of potable
water during the Phase 2 dredging. Before the end of the 2012 dredging season, EPA will make
a decision about whether to continue to pay for water full-time during the remaining off-seasons,
or to pay for water on a more limited basis during those off-seasons. EPA will make that
decision after additional data is gathered about the impact (if any) of the dredging activities on
off-season PCB levels in the Upper Hudson.
As a result of this decision, EPA is proposing to change the way it uses the 500 parts per trillion
(ppt) standard during Phase 2. Rather than treat the 500 ppt standard as the maximum allowable
Total PCB concentration in the water column (as measured at the far-field monitoring stations),
such that a confirmed exceedance of the standard would require a temporary halt of dredging
activities, EPA is proposing to use 500 ppt as a control level that will allow EPA to require GE to
Hudson River PCBs Site Page 10
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
propose and implement operational changes to reduce the PCB levels in the water column.
Additionally, EPA would also be able to require operational changes to reduce the level of
resuspension in the event of a seven-day running average of more than 350 ppt at one of those
monitoring stations. Such operational changes might include such measures as relocation of
dredge operations to lower concentration areas, reduction of the number of dredges working at
the same time, or temporarily moving dredges to lower velocity areas of the river. EPA would
also retain the discretion to require a temporary halt to dredging activities.
The primary reason for the inclusion of the 500 ppt standard in the Phase 1 Resuspension
Standard was to protect the water supply intakes at Waterford and Halfmoon (Statement of the
Engineering Performance Standards for Dredging (EPA, April 2004), Vol. 1, pp. 37, 93; Vol. 2,
p. 47). This reason no longer applies because the two towns are not going to use the Hudson
River as their source of potable water during the Phase 2 dredging.
Along with the daily PCB load standard, the 500 ppt and 350 ppt control levels would be used to
help monitor the amount of resuspension that is occurring and help signal whether operational
changes need to be made.
The conclusions of this Addendum support EPA's proposed changes to the Engineering
Performance Standards, and help demonstrate that the dredging operation can be modified in a
number of respects to obtain significant improvements over the Phase 1 performance in Phase 2.
EPA will closely review the data from the first year of the Phase 2 dredging and in light of that
data, as appropriate, consider further changes to the performance standards and/or operational
details for the second year of Phase 2. The Agency anticipates a similar adaptive management
approach to be followed during the remaining years of the project.
Hudson River PCBs Site Page 11
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
TOPIC 1
WATER COLUMN LOADS
DURING AND AFTER DREDGING
-------
TOPIC 1-A
PCB CONCENTRATIONS AT THE FAR-FIELD
STATIONS:
POST-DREDGING VS BASELINE
-------
Topic 1-A - PCB Concentration at the Far-Field Stations: Post-Dredging vs. Baseline
The purpose of this analysis is to compare Total PCB (TPCB) concentrations at the far-field
monitoring stations at Thompson Island Dam, Schuylerville (Lock 5), and Waterford during the
post-dredging period from December 1, 2009 through April 6, 2010 versus the baseline period
between 2004 and 2008. The goal is to discern whether dredging-related PCB release is evident
after completion of the Phase 1 activities or whether conditions related to PCB transport have
returned to baseline.
1-A.l. Post-Dredging Monitoring
Monitoring data collected by GE for the post-dredging period shows that the TPCB
concentrations measured at the far-field stations are comparable to TPCB concentrations
observed at these stations during the baseline period. Figures 1-A-la, 1-A-lb and 1-A-lc show
the water column concentrations for the post-dredging period, beginning December 1, 2009, for
the three far field stations, Thompson Island, Lock 5 (Schuylerville) and Waterford, respectively.
The plots also show the available baseline data for the period prior to dredging (2004 to 2008).
The range of baseline concentrations during the December to April period is denoted by the
region colored in gray on each plot. Figure 1-A-la shows the TPCB concentrations at the
Thompson Island station for all flows between the dates identified above, including two high
flow events from March 23 to 28, 2010 and April 1 to 4, 2010. Note that the amount of baseline
data for the Thompson Island station is limited to early December and late March. Additionally,
largely due to access limitations, there is very little high flow data for this station. Nonetheless,
the data shows that TPCB concentrations dropped to within baseline ranges soon after the
remedial activities stopped on December 4, 2009 and remained within the low-flow baseline
range except during the January, March and April 2010 high flow events. (While dredging and
capping/backfilling within the 10 Certification Units [CU] was completed on December 4, 2009,
demobilization of dock structures and related floats continued through December 23, 2009 and
one tug boat remained in the water after December 11, 2009). A similar observation can be made
for the Lock 5 and Waterford stations, as shown in Figures 1-A-lb and 1-A-lc, respectively.
Once again, the data shows that TPCB concentrations have fallen to within baseline ranges
except during high flows at Lock 5. At the Thompson Island and Lock 5 monitoring stations,
high flows are defined as flows greater than or equal to 10,000 cfs whereas at the Waterford
monitoring station high flows are defined as flows greater than or equal to 14,000 cfs due to
contributions from major tributaries (Hoosic River and the Batten Kill).
1-A.2. High Flow Monitoring
While it is clear that the water column concentrations returned to baseline concentrations under
low flow soon after all remedial activities ended, conditions under high flows are not as easily
characterized. In particular, post-dredging high flow monitoring data collected at the Thompson
Island monitoring station has yielded some of the highest observed concentrations in the Upper
Hudson. Notably, the values observed during the March and April 2010 high flow events were,
in some cases, more than an order of magnitude higher than the highest TPCB values observed
Hudson River PCBs Site Page 1-A-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
during dredging (e.g., 13,400 ng/L on March 25, 2010, us. 686 ng/L on October 14, 20091). The
cause of these post-dredging observations was unclear at first but the likely explanation became
clear when all five of the intakes were shown to be badly fouled with plant debris and perhaps
damaged. This was discovered during an April, 2010 maintenance inspection. It is believed that
the plant debris clogged the intakes as well as forced them into the river bottom during high
flows. One of the intakes was so badly fouled that it could not be found with an underwater
camera and had to be snagged with a grappling hook. All five intake screens were clogged with
plant matter and mud. Since the last routine maintenance visit was conducted in November 2009,
it is unclear when the intakes became fouled, but all data obtained since November 2009 must be
viewed with caution. Figure l-A-2a is a photograph showing an example of the fouling at three
of the Thompson Island intakes.
Similar but less extensive fouling was observed for the Lock 5 as shown in Figure l-A-2b. In this
instance, the debris screen for one intake was lost entirely and the intake itself was clogged with
plant matter and mud. For two other intakes, the routine backflush revealed that visible quantities
of mud had lodged in the tubes. The reported poorly operating pumps at Thompson Island and
pump failure at Waterford suggest that intake fouling impacted the ability of the sampling
systems to draw water in addition to compromising the properties of the water drawn. Related
sampling evidence is discussed below.
1-A.2.1. High Flow Monitoring: Comparison of Post-Dredging Period to Baseline
Period
In Figure l-A-3a, the TPCB concentrations at the Lock 5 station measured during the two post-
dredging high flow events from March 23 to 28, 2010 and April 1 to 4, 2010 are compared with
similar high flow data collected during all months in the baseline period. Post-dredging TPCB
concentrations at high flows at Lock 5 are close to, but still exceed similar measurements for the
pre-dredge period. In light of the discovery of the fouled intake and mud collection within the
intake lines, the results are considered suspect. As shown in this figure, the TPCB concentrations
measured in many of the samples from the March 23 to 28, 2010 high flow event are between
100 ng/L and 500 ng/L. Notably, the highest measured value during the March 2010 sampling
appears to be equivalent to the highest TPCB concentration observed at this station during active
dredging in 2009. Because of its similarity to the suspect samples at Thompson Island (discussed
further below), this sample is suspect as well. In direct contradiction of the March observations,
samples collected during the April 1 to 4, 2010 post-dredging high flow event appear to show
TPCB concentrations that are within the range observed during the baseline period (10 ng/L to
50 ng/L).
Figure l-A-3b presents the TPCB concentrations at the Waterford station measured during three
post-dredging high flow events from January 26 to 28, 2010, from March 16 to 29, 2010 and
from April 1 to 4, 2010 as compared with similar high flow data collected during all months in
the baseline period. (Note that three high flow events were observed at Waterford during the
post-dredging period, as opposed to only two such events at the two upstream stations.) Unlike
1 This value was obtained along with a split sample that yielded a concentration of 306 mg/L. The factor of 2
variation between these samples was typical of the poor sample replicate precision noted at Thompson Island during
dredging.
Hudson River PCBs Site Page l-A-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
the data from samples at the Lock 5 monitoring station, the TPCB concentrations in nearly all
post-dredging samples at Waterford station fall well within the pre-dredging baseline range for
high flow conditions (6 ng/L to 200 ng/L). Values for TPCB concentrations in three samples
from the March 16 to 29, 2010 high flow event are greater than or equal to the highest TPCB
concentration observed at Waterford after June 1, 2009 during active dredging in the Upper
Hudson River. Also, the highest value obtained during the March high flow event (1,900 ng/L) is
nearly an order of magnitude greater than the highest value observed post-June 1, 2009 (201
ng/L). As will be further discussed below, these data are suspect and most likely indicative of
fouling of the water intakes at the Waterford monitoring station. Other than these three samples,
water column concentrations at Waterford appear to have returned to baseline conditions under
both low and high flow. There is no evidence of dredging-related PCB loads after the remedial
activities were concluded.
1 -A. 2.2. Impact of Intake Fouling
The next set of three figures provides indirect chemical evidence of the impact of intake fouling.
The figures depict the TPCB concentrations normalized to TSS (i.e., ng TPCB/mg solids) versus
flow (Fort Edwards flow for the Thompson Island and Lock 5 stations and Waterford flow at the
Waterford station) measured at the three far-field monitoring stations during the post-dredging
period between January and March 2010. Since PCB transport during the high flow events is
generally considered the result of sediment resuspension due to higher water velocities, it is
reasonable to expect that this normalization should yield a consistent relationship between TSS
and water column concentration. In particular, given the broad integrating nature of resuspension
during high flow events (i.e., increased water velocities occur throughout the river during high
flows, increasing the rate of resuspension from sediments across much of the river bottom ), it is
reasonable to expect that the levels of PCBs on suspended matter should not be highly variable
during these events.
Figure l-A-4a shows that at Thompson Island, the normalized TPCB concentration ranges from
less than 1 to more than 1,000 ppm during higher flows (i.e., greater than 10,000 cfs). Also the
normalized concentration is notably higher during high flows as opposed to low flow conditions,
suggesting that the nature of the solids transported during high flow are different from those
transported at low flow. This is distinctly different behavior from what is observed at the
downstream stations. As noted previously, it is likely that the high concentration, high
normalized concentration samples are due to the plant fouling observed at all five intakes for the
station.
In contrast, Figure l-A-4b shows that at Lock 5, for all but one sample, the normalized TPCB
concentration is less than 20 ppm regardless of flow. Additionally, there is no substantive change
in the normalized concentration with flow, although the trend suggests there may be some
reduction in variability at higher flows. This relationship is much more consistent with the
expectation described above. It also contrasts sharply with the data from the Thompson Island
station, located only 6 miles upstream. However, as described previously, the Lock 5 station are
still suspect. One sample yielded a normalized PCB concentration that was more than 3 times
greater than the any other value. This result occurred in the sample with the highest reported
concentration at this station post dredging (560 ng/L), very close to the highest reported for the
station during dredging (643 ng/L). As described below, the occurrence of this sample did not
Hudson River PCBs Site Page l-A-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
coincide with any flow-related change that might have suggested a mechanism to cause it,
adding to its uncertainty. The other high flow samples fall outside the range of baseline high
flow data, which might suggest dredging-related PCB transport {i.e., PCB redistribution) on top
of the normal baseline loads. However, this observation is not borne out by similar above
baseline transport at Waterford and so is considered suspect in light of the compromised intakes.
The normalized TPCB results for the Waterford station show a similar relationship to that at
Lock 5. The results presented in Figure l-A-4c indicate that, except for three samples, the
measured normalized TPCB concentration is less than 5 ppm during higher flows.2 Like the
samples at Lock 5, the results show a fairly consistent relationship across the range of flows,
again with the suggestion of reduced variability in normalized TPCB at the highest flows. Also
similar to the Lock 5 results, the three samples with unusually high normalized TPCB
concentrations are also the three with highest absolute TPCB concentrations. The normalized
TPCB levels in these samples are nearly an order of magnitude higher than any other sample
obtained under similar high flow conditions (high flows at Waterford are defined here as greater
than 14,000 cfs). These samples are also without corresponding hydrologic changes and appear
unrelated to conditions at Lock 5 based on time-of-travel considerations, described further
below.
Given the direct evidence for the intake fouling at Thompson Island and Lock 5, neither the
March nor April 2010 high flow results for these stations can be considered accurate
measurements of load in the water column during this period. The observation that the fouling
was less extensive at Lock 5 than at Thompson Island suggests that the Lock 5 reported values
are perhaps less compromised than those at Thompson Island. Thus the Thompson Island
samples show no consistency in their normalized TPCB levels whereas the Lock 5 samples
generally have consistent results. Notably, nearly all Waterford high flow samples fall within the
range of baseline results (described below). These observations suggest that the extensive fouling
at Thompson Island caused the intakes to incorporate muds directly from the river bottom,
thereby changing the normalized TPCB levels during high flow. The less compromised Lock 5
station simply accumulated water borne sediments that were later incorporated in samples. The
results for Lock 5 suggest that there may be some evidence for dredging-related PCB transport
{i.e., PCB redistribution) at this station but the actual amount of transport cannot be quantified.
The high flow samples in March 2010 are at best upper bound estimates and are not considered
reliable. Taken together, these observations indicate that there is no direct evidence of dredging
related PCB transport to the Lower Hudson after completion of the dredging activities in
December 2009, and some suggestive Upper Hudson transport was seen during only one event
(March 23-28, 2010) at only one station.
2 The lower average normalized TPCB value is consistent with a conceptual model of load originating primarily in
the Thompson Island Pool. Transit downstream adds water and additional, less contaminated sediments, while
transporting the PCBs fairly conservatively. This behavior is probably best exhibited during high flow events where
loads remain the same or increase and downstream transport occurs rapidly. Constant or increasing loads
downstream were also observed during baseline monitoring.
Hudson River PCBs Site Page l-A-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
l-A.2.3. PCB Concentrations during March 2010 High Flow Event
During the period of high flows in March and April 2010, the time of travel between Thompson
Island and Lock 5 (Schuylerville) was on the order of 4 to 6 hours. There are no major tributaries
entering the river between the stations but the TSS load increases from Thompson Island to Lock
5, suggesting resuspension of sediments in the intervening 6 miles. (The Batten Kill enters the
Hudson just below the Lock 5 station.) Despite the increasing TSS load and short travel time,
overall PCB concentrations and PCB load declined markedly during this high flow period. This
is shown in Figure l-A-5a, which compares flow and concentrations over time at Thompson
Island and Lock 5.
Evident in the figure is the rapid variation in concentration at the Thompson Island station, in
contrast to the much more gradually varying concentrations at Lock 5. Other than the second
sample obtained at Thompson Island, the higher concentration samples occur erratically during
declining flows. Additionally, the concentrations observed during the first 48 hours of the
sampling period are as much as 20 times greater than the highest concentrations observed during
the dredging operation itself, despite the lack of dredge plumes, oil sheens and other factors that
were present during dredging but absent during this post-dredging study period.
Note that through most of this period, the samples represent 6-hour composites collected
continuously one after the other. Given the relatively smooth hydrologic record observed during
this period, sequential samples are expected to be similar to each other. This was, in fact,
observed fairly consistently at Lock 5 but clearly not at Thompson Island.
Note as well the lack of correlation between the samples at the 2 stations. None of the high
concentration samples at Thompson Island result in a similar response downstream, save one
sample collected at 1800 hours on March 25, 2010. Even in this instance, the single high value at
Lock 5 actually predates by 4 to 6 hours the arrival of the putative 13,600 ng/L water from the
Thompson Island station. Both the lack of correlation between the stations and the rapid apparent
rate of TPCB concentration decline between stations indicate that the results of the two stations
are inconsistent. Given the particularly unusual TPCB behavior at the Thompson Island station
and the documented intake fouling at both stations, the lack of correlation between station results
is not surprising. Of the two, the Thompson Island data are considered to be the least related to
actual in-river conditions but data from both stations are subject to considerable uncertainty.
As noted previously, the one high concentration sample at Lock 5 obtained at 1800 hours on
March 25, 2010 is inconsistent with all of the other samples obtained at this station during the
January to March 2010 post-dredging period, based on the normalized TPCB concentration.
When viewed as part of the sample collection sequence, it is also inconsistent with the observed
trend. Neither adjacent sample gives any suggestion of the short-term change in absolute or
normalized concentration shown by this sample. As a result, this sample is considered unrelated
to the true river conditions during this time.
Figure l-A-5b presents a similarly constructed comparison for the Lock 5 and Waterford
stations, with the flow at Waterford shown instead of the flow at Ft Edward. The majority of the
sample results at both stations appear to vary smoothly and similarly over time. The suspect
sample at Lock 5 is noted on this diagram with a magenta circle. Three Waterford samples are
also noted in this manner, representing the three samples discussed earlier, with high normalized
Hudson River PCBs Site Page l-A-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
TPCB levels. The highest of the samples appears to be an outlier relative to its adjacent samples,
much as the one sample at Lock 5. However, the other two identified samples at Waterford occur
sequentially, suggesting they may be related, although they are clearly unlike the majority of
Waterford samples. As it turns out, the two samples follow a 24 hour period marked by a pump
failure, in which no samples were obtained at the station. While the complete details have not
been provided to EPA as of this writing, it is suspected that sediment accumulation in the lines or
simply fouling of the intake at Waterford is the likely cause of the suspect sample results. For the
reasons given above, the three identified samples at Waterford are not considered representative
of in-river conditions at the time of their collection.
While the lack of correlation as well as the chemical properties clearly identify the Thompson
Island station samples as well as a few downstream samples as suspect, it is useful to examine
PCB loads during the post-dredging period to form a further basis for their dismissal. EPA
compared the TPCB loads at the three far-field monitoring stations during Phase 1 dredging
(May 15 to November 30, 2009) as well as for the main post-dredging high flow event (March
23 to 28, 2010). These TPCB loads are shown in Figure l-A-6. The left panel shows the TPCB
loads during active dredging and the right panel shows the corresponding loads during the March
23 to 28, 2010 post-dredging high flow event.
For the left panel, the cumulative total load is shown for each station, broken down into the
estimated baseline contribution, shown in blue, and that due to dredging, shown in green. Each
bar is also annotated to show the overall level that corresponded to 117 kg over baseline as well
as the level corresponding to 1 percent of the mass removed. Note that the load at Waterford did
not exceed this threshold.
A decline in the TPCB transport from Thompson Island to Lock 5 is evident in both panels.
While the decline during dredging may be questionable (see the EPA Phase 1 Evaluation Report,
Chapter 2)3, it is clear that the rate of decline during dredging is significantly less than during the
post-dredging high flow event. The loads at each station for Phase 1 were obtained from the
simple product of daily data and flow for the May to November 2009 period. The loads for the
high flow event were also obtained as the simple product of flow and concentration for the
period. Loads during the high flow event at Lock 5 and Waterford are calculated with and
without the suspect data. The suspect data have only a small effect at Lock 5 but serve to more
than double the estimated total load at Waterford.
In both panels, the load decline from Thompson Island to Lock 5 is noted. In particular, the load
decline during dredging is quite substantial, at 30 percent of the total transport. This decline
occurs largely during low flow conditions. While there are concerns over the Thompson Island
data, some of this decline may be the result of settling or gas exchange losses during transit to
3The results obtained at the Thompson Island station were also problematic during dredging. In particular, when
higher concentrations were obtained at this station, they were not consistently observed downstream at Lock 5, as
might be expected. Additionally, replicate samples at the Thompson Island station typically exhibited poor
reproducibility when concentrations approached or exceeded 500 ng/L. The presence of oil sheens were frequently
observed during Phase 1 and had the potential to confound sample results at this station. These observations have
potentially compromised the use of this station data for Phase 1 calculations. As a result, the data from Thompson
Island during Phase 1 are undergoing thorough review at the time of this writing.
Hudson River PCBs Site Page l-A-6
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
the Lock 5 station. Flows are generally low during this period, so as to provide 12 to 18 hours of
time for these losses to occur. This decline is dwarfed, however, in comparison to the loss that
occurs during the five day high flow event. Nearly 90 percent of the apparent load is lost during
a period of rapid transit (4 to 6 hours) while gaining TSS load between the stations (not shown
here). It is this observation that finalizes EPA's perspective on the Thompson Island data. This
loss is simply too unlikely and without a probable mechanism to be an actual representation of
in-river conditions. For all of these reasons, the Thompson Island post-dredging high flow data
are considered suspect. As mentioned previously, given the extent of fouling observed, all
Thompson Island post-dredging samples must be viewed with caution.
The load data at Lock 5 and Waterford are also telling here. While the following analysis does
not provide the same strength of evidence as Thompson Island, it can be seen that by eliminating
the 4 suspect values at Lock 5 and Waterford based on their normalized TPCB levels, the loads
can be shown to match fairly closely, with about a 5 to 10 percent increase in load between the
stations. However, the scale of this load gain is lower than historical observations between the
stations during baseline high flow events. This observation suggests that some component of the
load at Lock 5 may be the result of dredging-related PCB transport. Given the uncertainty of the
sample results at Lock 5, it is not clear what the actual flux of dredging-related PCB would be.
However, the load as calculated is quite small relative to the active Phase 1 period and includes
the baseline load as well. (The cumulative load at Lock 5 was about 350 kg including baseline,
see Figure l-A-6) Perhaps, more importantly, the high flow event that followed in early April
showed concentrations consistent with baseline, suggesting no further dredging-related PCB
transport.
To complete its review of the post dredging monitoring, EPA examined the "fingerprint" of the
PCBs being transported by the river before, during and after Phase 1. This was accomplished by
examining the Tri+ fraction of the PCBs in the water column. This ratio is a fairly sensitive
indicator of the extent of dechlorination on the PCB mixture and has served well in the past as a
diagnostic tool (e.g., EPA, 1997. Hudson River PCBs Phase 2 RI/FS - Data Evaluation and
Interpretation Report). In Figure l-A-7, EPA has plotted the Tri+ to TPCB ratio versus the
TPCB concentrations observed at the Waterford station during three periods: baseline (2004 to
2008); active dredging (May 15 to November 30, 2009) and post-dredging (December 1, 2009 to
March 29, 2010). A careful examination of this figure shows that the patterns in the upper panel
(baseline) and the lower panel (post-dredging) are similar to each other and different from the
pattern observed in the middle panel (active dredging). Specifically, high concentrations are
marked by a high fraction of Tri+ PCB. This is attributed to scouring of relatively low
concentration, less dechlorinated PCB mixtures present at the sediment surface. In contrast, the
relatively high concentrations and high loads observed during dredging were driven by the
resuspension of the more concentrated, more dechlorinated sediments that were associated with
the sediment removed by dredging.4
4 As extensively documented in the Hudson River PCBs Phase 2 RI/FS - Data Evaluation and Interpretation Report
(EPA, 1997), the extent of dechlorination increases with PCB concentration, thus marking the most highly
contaminated sediments as most dechlorinated.
Hudson River PCBs Site Page l-A-7
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
This demonstrates that the PCB mixture in the water column of the Upper Hudson River during
the post-dredging period has returned to the pattern observed prior to the Phase 1 dredging event
in 2009. Since one of the remedial goals of the ROD was to reduce PCB mass in the Upper
Hudson, the remedy targeted the concentrated deposits of PCB mass. Thus the Tri+ fraction in
PCBs released by dredging is characteristically lower than that observed during baseline
conditions. It is clear from the evidence presented in the third panel in Figure l-A-7 that the
concentrations and Tri+ fraction have returned to the baseline relationship. Contrary to what GE
has asserted in its Phase 1 Evaluation Report, EPA believes that the data collected by GE
actually shows no significant short-term impacts to the water column of the Upper or Lower
Hudson River due to the redistribution of PCBs from dredging.
1-A.3. Summary
This analysis has identified the following concerns:
• PCB concentrations at the Thompson Island station post-dredging must all be considered
suspect, particularly those collected during recent high flow events. Documented intake
fouling as well as a geochemical analysis of the nature of the PCB concentrations and
patterns indicate that these results are not indicative of in-river conditions.
• Samples from Lock 5 and a limited number of samples at Waterford stations obtained during
the post-dredging period are also clearly suspect, based on based on evidence of fouling at
Lock 5 as well as geochemical lines of evidence at both stations.
• The inability to maintain the automated sampling apparatus over the winter from the physical
limitations due to ice cover and safety is likely to have compromised much of the post-
dredging data to date.
• Changes in PCB "fingerprint" that might indicate the presence of dredging related PCBs are
only shown to occur in samples that appear suspect based on the ratio of PCBs to suspended
matter. The inconsistent and random occurrence of these samples during the post-dredging
monitoring period as well as the evidence of fouling and pump problems suggest these data
are unreliable measures of in-river conditions.
• Nonetheless, a large number of the samples obtained at the Lock 5 and Waterford stations,
and even some of the Thompson Island samples are consistent with the PCB concentrations
and patterns observed prior to Phase 1. These samples occur fairly consistently over the post-
dredging period, lending credence to their interpretation and suggesting the absence of
dredging-related PCB transport during these sampling periods.
• A limited number of samples at Lock 5 suggest a single, short term transport event of
dredging-related PCBs, but do not provide an acceptable basis to quantify this transport.
Taken at face value, the results indicate only a small transport event.
• Estimates of load to the Lower Hudson as measured at Waterford during high and low flow
conditions in the post dredging period show no increase over baseline loads. That is, the
relationship between flow and concentration observed during the post dredging period at
Waterford is the same as that observed prior to Phase 1.
• The post-dredging evidence collected to date shows no significant short-term impacts to the
Upper or Lower Hudson River due to the redistribution of PCBs from dredging.
Hudson River PCBs Site Page l-A-8
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
100,000
10,000
O)
I 1,000
CO
!_
c
CD
o
c
o
O
GO
cl 100
ro
o
~ Baseline (2004-2008, Dec. - Apr.)
T Post-Dredging (Low Flow)
• Post-Dredging (High Flow)
(Mar. 23 -28)
¦ Post-Dredging (High Flow)
(Apr. 1 - 4)
V T
10
~
~
Water column concentrations
have fallen to within baseline
ranges other than at high
flows.
w
#•
»
x
~~
Note:
Non-detects are marked with an X and reported as 1/2 detection limit.
Range of available
December to April
baseline samples.
I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I | I I I I I I
comc\iCT>mc\iCT>coc\iCT>tococ\ia>tocooto
CsJ CsJ
CNJ CNJ CNJ
Month/Day
eg eg
CO CO CO
Post-Dredging Monitoring at Thompson Island
Total PCB Concentration vs. Date
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-A-la
April 2010
-------
1,000
g 100
c
o
CO
-t—'
c
CD
O
c
o
O
m
o
CL
ro
-«—<
o
~ Baseline (2004-2008, Dec. - Apr.)
~ Post-Dredging (Low Flow)
• Post-Dredging (High Flow)
(Mar. 23-28)
¦ Post-Dredging (High Flow)
(Apr. 1 -4)
Note change in vertical scale.
Water column concentrations
have fallen to within baseline
ranges other than at high
flows.
10 -
*
~ * ~ V
~ ~~~
A
%
X
>^<
• •
~ ~ ~
~ ~
~ A
Range of available
December to April
baseline samples.
Note:
Non-detects are marked with an X and reported as 1/2 detection limit.
1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111
C\J C\J
1 1 II
LO
III
C\J
i i |
a>
T—
CNj
CNj
C\J
C\J
C\J
I I I I I I I I
coLoc\ja>Loc\ja>coc\ja>cococ\ja>cocooco
£N
N CM CO CO CO
C\J CO
Month/Day
Post-Dredging Monitoring at Lock 5 (Schuylerville)
Total PCB Concentration vs. Date
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-A-lb
April 2010
-------
10,000
1,000 -
O)
c
o
CD
i—
c
CD
c 100
o
O
GO
O
CL
ro
o
~ Baseline (2004-2008, Dec. - Apr.)
T Post-Dredging (Low Flow)
~ Post-Dredging (High Flow)
(Jan. 26-28) "
• Post-Dredging (High Flow)
(Mar. 16-29)
¦ Post-Dredging (High Flow)
(Apr. 1 - 4)
Note change in vertical scale.
Water column concentrations
have fallen to within baseline
ranges other than at high
flows.
10
* * *
~ ~~
Range of available
December to April
^ baseline samples.
1 —i m 11111 m 1111111111111111111111 m | m 1111111111111111111111 m 11111 m 1111111111111111111 m 1111 m 11111111111111111111111111111
COLOCMCDLOCMCDtOCMCDtOCOCMCDtOCO
eg eg
eg eg
eg eg eg
eg
eg eg
Month/Day
^ eg
eg eg
CO CO
eg co
CO CO CO
CD
Post-Dredging Monitoring at Waterford
Total PCB Concentration vs. Date
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-A-lc
April 2010
-------
TID Intake Screen #3 being pulled from river and covered with weeds and mud
TID Intake #1 covered with
weeds and mud
Thompson Island Intake Photos
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1 -A-2a
April 2010
-------
-------
1000
CO
c
c
o
CO
-t—'
c
CD
o
c
o
o
m
o
CL
ro
-*—>
o
100 -
Suspect Sample
See text for discussion
fe".
~
~ •
10
1
10,000
~
~
I I I I I I
Max i niu nivaju_e_qbseryed at_Lock
5 during active dredging.
Post dredge
concentrations at high
flows at Lock 5 are close
to, but still exceed similar
high pre-dredge high
flow events.
Baseline (2004-2008,
flow> 10,000 cfs, all months)
Post-Dredging High Flow
(Mar. 23 - 28)
Post-Dredging High Flow
(Apr. 1 - 4)
i
15,000
1 I 1 r"
20,000
Flow at Ft Edward (cfs)
i
25,000
30,000
High Flow Monitoring at Lock 5 (Schuylerville)
Total PCB Concentration vs. Fort Edward Flow
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-3a
April 2010
-------
10,000
1,000
CD
c
o
CC
-i—•
c
0
c 100
o
O
m
o
CL
ro
-*—>
o
10
_l I I L_
0
Baseline (2004-2008,
flow> 14,000 cfs, all months)
Post-Dredging High Flow
(Jan. 26-28)
Post-Dredging High Flow
(Mar. 16-29)
Post-Dredging High Flow
(Apr. 1 - 4)
Suspect Samples
See text for discussion
Maximum value observed at
Waterford duringiactive dredging.
X
% **
~ ~
~
~
~
*
~%
~
^ Maximum value
observed at
Waterford post-
June 1 during
active dredging.
Unlike the Lock 5 station, concentrations at Waterford fall well
within the pre-dredging baseline range for high flow conditions for
nearly all post-dredging measurements.
~i 1 r
~i 1 1 r
~i 1 1 r
~i 1 r
10,000
20,000 30,000
Flow at Waterford (cfs)
40,000
50,000
High Flow Monitoring at Waterford
Total PCB Concentration vs. Waterford Flow
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-3b
April 2010
-------
10"
1000 -
o
cn
CD
E
m
o
CL
CD
C
CO
C/3
H
m
o
CL
ro
-*—>
o
100 -
10 -
1 -
Thompson Island
_l I I I I I I I I I I I I I I I I I l_
a Total PCB Cone <500 ng/L
a Total PCB Cone between 500 to 1000 ng/L
a Total PCB Cone >1000 ng/L
~
~
I
~
~
i
A
~
~
At higher flows, total PCB
concentration ranges from 10
to 1,000 ppm.
0.1
Data from Jan-Mar 2010
25,000
Ft. Edward Flow (cfs)
Total PCB Concentrations Normalized to TSS Thompson Island
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-4a
April 2010
-------
100
• Total PCB Cone <500 (ng/L)
• Total PCB Cone, between 500 to 1000 ng/L
-Suspect Sample
See text for discussion
o
cn
CD
m
O
CL
CD
en
H
m
o
CL
ro
-*—>
o
10 -
1 -
0.1
0
For all but 1 sample, total PCB
concentration is less than 20 ppm.
~r i i i 1 t t i i | i i r
5,000 10,000
Data from Jan-Mar 2010
T
~i 1 1 1 1 1 1 1 r
15,000
T
20,000
25,000
Ft. Edward Flow (cfs)
Total PCB Concentrations Normalized to TSS Lock 5
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-4b
April 2010
-------
o
CO
CD
E
m
O
CL
CD
c
CO
C/3
H
m
O
CL
ro
-*—>
o
Waterford
100
_l I I I I I I I I I I I I I I I I I l_
_l I I L_
_l I L_
Total PCB Cone. <500 ng/L
Total PCB Cone. >1000 ng/L
Data from Jan-Mar 2010
Suspect Samples
See text for discussion
10 -
For all but 3 samples, total PCB concentration is less than 5
ppm. Most are less than 1 ppm.
1 -
0.1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 r
5,000 10,000 15,000 20,000 25,000 30,000 35,000
Waterford Flow (cfs)
Total PCB Concentrations Normalized to TSS Waterford
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-4c
April 2010
-------
14,000
13,000
3,000
CO
c
c
o
CO
2,500 -
m
O 1,500
CL
ro
o
1,000 -
500
0
3/23
H Composite
0 Grab
¦ TID
1 | Lock 5
3/24
3/25
3/26 3/27
Start Date/Time
3/28
3/29
25,000
- 20,000
- 15,000
5,000
3/30
Thompson Island and Lock 5 Total PCB Concentration and Fort Edward
Flow vs. Date
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-5a
April 2010
-------
3/23
3/24
3/25
3/26 3/27
Start Date/Time
1 r
3/28
3/29
40,000
- 35,000
- 30,000
- 25,000 ^
CO
CD
o
- 20,000 s.
T]
O
15,000 a
' O)
10,000
- 5,000
3/30
Lock 5 and Waterford Total PCB Concentration and Fort Edward
Flow vs. Date
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-5b
April 2010
-------
600
500
400
~u
nj
O
_l
m
O 300
Q.
ro
o
200
100
Total: 530
Thompson Island
Total PCB Load During Dredging
(5/15/2009 -11/30/2009)
Lock 5
Far-field Stations
¦ Net Cumulative Load
¦ Baseline
1 % of mass removed
117 kg over baseline
Waterford
Thompson Island
Total PCB Load at Far-field Stations
Mar. 23 - Mar. 28,2010
Lock5
Far-field Station
~ TPCB Load as Measured
¦ TPCB Load as Corrected
Waterford
Comparison of Total PCB Load During Dredging vs. Post Dredging
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-A-6
April 2010
-------
1 1 1 r
300 350
n i | i i i i | i i i i | i i i i | i r
100 150 200 250
Total PCB Concentration (ng/L)
During Dredging (5/15 -11/30/2009)
350
Total PCB Concentration (ng/L)
1
0.9
Post Dredging (12/1/2009 - 3/29/2010)
J I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I L
% 0.8
-------
TOPIC 1-B
NEAR-FIELD PCB RELEASE
MECHANISM STUDY
-------
Topic 1-B - Near-Field PCB Release Mechanism Study
1-B.l. Objectives and Field Implementation
The Near-Field PCB Release Mechanism Study was conducted by GE to assess the nature of the
primary release mechanism in the vicinity of dredging operations. The Data Quality Objectives
(DQOs) of this study were to evaluate the extent to which the PCBs released by remedial
operations are dissolved or associated with suspended matter and to determine whether near-field
TSS concentrations can be a reliable indicator of PCB releases. This study is described in the
Remedial Action Monitoring Plan Quality Assurance Project Plan (RAM QAPP; GE, 2009). The
RAM QAPP specified that the near-field PCB release mechanism study was to be conducted in
five areas so that a range of dredging conditions could be evaluated, including: different
sediment types (cohesive and non-cohesive), sediment PCB concentration ranges, and the range
of anticipated dredge types). However, due to project logistics, dredging was not performed in
four of the five areas designated in the RAM QAPP for this study. As a result, the study was
conducted only in the East Griffin Island Area (EGIA).
Dissolved and particulate samples were collected in the following locations in the vicinity of
EGIA dredging activities.
• EGIA100M-UP (point; 100 m upstream of the dredging operation)
• EGIA30M-DN (transect; 30 m downstream of the dredging operation)
• EGIA100M-DN (transect; 100 m downstream of the dredging operation)
• EGIA300M-DN (transect; 300 m downstream of the dredging operation)
Monitoring locations are shown on Figure 1-B-l.
Water samples were filtered in the field at the time of collection, and the filters and the filtered
water submitted for analysis (GE, 2010).
1-B.2. Nature of PCB Release
Based on the results obtained from study, both EPA and GE in their March 2010 reports
observed that PCBs in the water column were predominantly in the dissolved form, averaging
over 90% of the total PCBs (Table 1-B-l). Furthermore, the dissolved PCB fraction was
dominated by mono- and di- chlorinated congeners, while the particulate fraction is dominated
by tri- and tetra-chlorinated congeners (EPA, 2010 and GE, 2010).
In addition, PCB-bearing oils were also released during dredging. This NAPL was observed as
small droplets of oil in the river downstream of the dredging operation and sheens occurred
frequently in the ERI and EGIA. Sheens were sampled on several occasions to assess PCB
concentrations. Duplicate samples showed large disparities probably due to variability in sheen
thickness and surface coverage and the amount of underlying water in the sample. However, the
oils were not isolated for analysis such that a specific congener pattern could be identified. EPA
believes that the presence of oil was partially responsible for the high degree of variability
observed in sample replicates when concentrations of PCBs in river water approached the 500
ng/L threshold. GE has argued that the filters would have removed the oils and that they did not
see much oily sediment in the processing facility.
Hudson River PCBs Site Page 1-B-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
GE has assumed that the oil droplets would be trapped on the filter when samples are filtered for
dissolved and particulate PCB analysis and that they would primarily influence the measurement
of particulate PCBs. Based on this assumption GE has concluded that an oil phase can at most
account for a minority of the PCBs measured at the far-field stations. However, this assumption
is not valid because such filters are not designed to retain liquids and the size (longest
dimension) of a typical PCB molecule with at least two chlorine atoms in meta- or ortho-
substitution positions is approximately 1.1 nanometers whereas the nominal pore size (diameter
of opening) for such filters is approximately 700 nm. The PCB-bearing oils will pass through the
filter and will become part of the dissolved PCBs sample and may in fact account for a
significant portion of the PCBs measured at far-field stations.
1-B.3. Use of TSS as an Indicator of PCB Releases
As stated above, one of the DQOs of this study was to determine whether or not near-field TSS
concentrations are a reliable indicator of PCB releases in the vicinity of remedial operations. For
the impact of dredging on a particular parameter to be assessed, it must first be shown that
variation in the parameter can be attributed to dredging. Evaluation of the PCB and TSS data
collected by GE on July 14, 15, and 17, 2009 at EGIA near-field stations indicates that the
release due to dredging at the time of sample collection cannot be accurately quantified.
In order to show that variation in PCB and TSS concentration is due to dredging, the same parcel
of water must be sampled at each near-field location, beginning with the upstream station {i.e.,
EGIA100M-UP) and ending with the southernmost transect (i.e., EGIA300M-DN). Using daily
velocities obtained from GE for the study area, approximate travel times were estimated between
each station on July 14, 2009 as follows:
• EGIA100M-UP to EGIA30M-DN: 36 min
• EGIA30M-DN to EGIA100M-DN: 9 min
• EGIA100M-DN to EGIA300M-DN: 12 min
The travel times on July 15 and July 17, 2009 were similar to those calculated for July 14, 2009.
All calculated average velocities and travel times are shown on Table l-B-2.
Examination of GE sample times on these dates showed that water samples were not collected on
the above schedule. The approximate time between GE sample collection events on July 14,
2009 were as follows:
• EGIA100M-UP and EGIA30M-DN: 4 hrs
• EGIA30M-DN and EGIA100M-DN: 2 hrs
• EGIA100M-DN and EGIA300M-DN: 1 hr
All GE sample collection times and differences are shown on Table 1-B-l. The lag in sample
collection times between the EGIA100M-UP and EGIA30M-DN and EGIA30M-DN and
EGIA100M-DN did reduce on July 15 and 17, 2009, but only to approximately 1 hour between
samples - a value still nearly double the calculated travel time between the EGIA100M-UP and
EGIA30M-DN stations and 5-6 times greater than the travel times between the other stations.
This information is summarized below. As the GE samples do not represent values obtained
from the same parcel of water as it moves downstream, the total and dissolved phase
Hudson River PCBs Site Page l-B-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
concentrations obtained at each station cannot be compared to evaluate PCB releases due to
dredging. Dissolved phase concentrations and total PCB concentrations change rapidly in the
near-field because of the dynamic nature of the dredging operation and related activities.
Therefore, the importance of collecting a sample from the same parcel of water as it travels
downstream cannot be overemphasized.
Comparison of Travel Time and Sampling Interval for Near-field Release Mechanism Study
Station
July 1
2009
July 15,2009
July 17, 2009
Travel
Interval
Travel
Interval
Travel
Interval
Time
between
Time
between
Time
between
(minutes)
samples
(minutes)
(minutes)
samples
(minutes)
(minutes)
samples
(minutes)
EGIA100M-UP to
36
226
36
62
46
80
EGIA30M-DN
EGIA30M-DN to
9
123
9
62
11
70
EGIA100M-DN
EGIA100M-DN to
12
65
12
62
15
72
EGIA300M-DN
Further, the samples were collected slowly, over a one-hour period of time, further compounding
the discrepancy between the actual travel time between stations and the time of GE sample
collection. For example, in order to capture the same parcel of water at both the EGIA100M-UP
and EGIA30M-DN stations, the timing of sample collection at both stations would need to
overlap, with collection EGIA30M-DN starting approximately 36 minutes after initiation of
collection at the EGIA100M-UP sample. Addition of the single hour between sample collection
events achieved on July 15 and 17, 2009 places the initiation of collection at the EGIA30M-DN
station approximately 30 minutes after the last of the 1-hour parcel of water sampled at
EGIA100M-UP has already passed.
This conclusion is further supported by the actual TSS and dissolved and particulate phase PCB
concentrations measured on July 14, 15, and 17, 2009. Neither TSS nor dissolved or particulate
PCB concentrations collected downstream of the dredging operation show trends of consistency
between sampling locations. TSS concentrations measured on July 14, 2009 fluctuated between
stations, increasing from 1.1 mg/L at the EGIA100M-UP station to 5 mg/L at the EGIA30M-DN
station and then declining to 1.4 mg/L at the EGIA100M-DN station, a value similar to that
observed upstream of dredging. TSS concentrations then increased again, reaching 2.4 mg/L at
the EGIA300M-DN station.
On July 15 and 17, 2009, TSS levels did increase in the downstream direction. However, on July
15, 2009 the total gain in TSS between the EGIA100M-UP and EGIA300M-DN stations was 1
mg/L. On July 17, 2009, the total gain between those locations was 5.6 mg/L. It would be
expected that other factors (such as river flows, effectiveness of resuspension controls, dredging
cycle times, presence of debris, proper bucket closure, boat traffic, and release of bucket decant
water) being the same, the largest increase in TSS above upstream levels would correspond to
the date with the greatest removal volume of sediments. This was not the case for this subset of
Hudson River PCBs Site Page l-B-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
days. Approximately 890 CY of sediment was removed from CU-17 on July 14, 2009, 456 CY
on July 15, 2009, and approximately 370 CY on July 17, 2009.
Dissolved phase PCB concentrations also fluctuated between stations on July 14, 2009,
increasing from 248 ng/L at the EGIA100M-UP station to 1,568 ng/L at the EGIA30M-DN
station, and then declining to 253 ng/L at the EGIA100M-DN station, a value similar to that
observed upstream of dredging. Dissolved phase PCB concentrations then increased again,
reaching 813 ng/L at the EGIA300M-DN station. PCB concentrations observed on July 15, 2009
displayed a similar pattern with concentrations of 448 ng/L, 1070 ng/L, 964 ng/L, and 1355 ng/L
measured from the northernmost to southernmost monitoring locations.
Dissolved phase PCB concentrations measured on July 17, 2009 increased from upstream of
dredging to downstream of dredging, with values of 839 ng/L, 973 ng/L, 3970 ng/L, and 8,220
ng/L from the northernmost to southernmost stations, respectively. Dissolved phase PCB
concentrations change rapidly in the near-field because of the dynamic nature of the dredging
operation and related activities. Therefore, the importance of collecting a sample from the same
parcel of water as it travels downstream cannot be overemphasized. Further, results of some of
the downstream dissolved phase samples may have been affected by the oil sheens that were
observed near CU-18.
Particulate phase PCB concentration patterns observed on July 14, 15, and 17, 2009 were similar
to those displayed by the dissolved phase. Particulate phase results from July 14 and 15, 2009
fluctuated in the downstream direction, with respective station PCB concentrations of 10.9 ng/L,
182 ng/L, 58 ng/L, and 95 ng/L observed on July 14, 2009 and 17 ng/L, 81 ng/L, 53 ng/L, and 98
ng/L measured on July 15, 2009. PCB concentrations generally increased in the downstream
direction on July 17, 2009, with respective station values of 14 ng/L, 33 ng/L, 275 ng/L, and 220
ng/L.
It should be noted that the EGIA100M-UP station is a single node whereas all other stations are
transects, with collection nodes spaced across the width of the river. Data obtained from transects
at the top and bottom of the EGIA during the near-field transect study (described in GE Phase 1
Evaluation Report Section 5.1.1) indicate that constituent concentrations varied across the
channel, both when the water entered the pool and when it left. This suggests that the river may
not be mixed enough for a sample collected from a single node from EGIA100M-UP to be
representative of baseline to the EGIA area.
1-B.4. Kinetics of PCB Release
GE has applied a two -phase equilibrium partitioning to the near-field release mechanism data
and concluded that the partitioning behavior was similar to that observed under baseline
conditions, indicating that there was no evidence of oil in the river. However this analysis cannot
be supported because: 1) a two phase partitioning is not valid in the presence of NAPLs, and 2)
the initial total mass of PCBs that released the PCBs to the dissolved phase is not known. The
total amount of dissolved phase PCBs added to the water column due to dredging is equal to the
sum of the following:
1. The resuspension and release from sediments at the dredged head
Hudson River PCBs Site Page l-B-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
2. Release of pore water due to impact of the dredge bucket on the sediment surface (bottom
disturbance), and
3. Any releases resulting from bucket decanting.
The presence of NAPL strongly suggests that the two phase partitioning model cannot be used to
discern the dynamics of PCB release in the near-field. In general, PCB release to the dissolved
phase is kinetically controlled because it takes longer for equilibrium partitioning to be achieved.
It has been reported that PCB congeners containing up to four chlorines approach equilibrium
within 6 weeks and congeners with greater than 6 chlorines may require months or years to reach
equilibrium (Coates and Elzerman 1986).
1-B.5. Conclusion
The near-field observations indicate that PCBs were dominated by the dissolved phase
component. Because the majority of the resuspended particles are deposited closer to the dredge
head, the dissolved PCBs they release is a small component of the dissolved phase
concentrations observed in the near-field water column. It is likely that the high dissolved phase
concentrations observed are due to the presence of NAPLs, which were constantly reported in
the EGIA area where the near-field release mechanism study was conducted.
References
Coates, J. T., and Elzerrnan, A. W. (1986). "Desorption kinetics for selected PCB congeners
from river sediments," Journal of Contaminant Hydrology 1, 191-210.
GE, 2009. Remedial Action Monitoring Quality Assurance Project Plan. Prepared for General
Electric Company, Albany, NY by Anchor QEA, Environmental Standards, Inc, and
Arcadis. May 2009.
GE, 2010. Phase 1 Evaluation Report. Prepared for General Electric Company, Albany, NY by
Anchor QEA, LLC and Arcadis. March 2010.
EPA, 2010. Phase 1 Evaluation Report. Prepared for EPA Region 2 and USACE Kansas City
District by The Louis Berger Group, Inc. March 2010.
Hudson River PCBs Site Page l-B-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 1-B-l
GE Sample Time Differences
Sample Collection
Sample Collection
Sample Collection
Collection Time
Collection Time
Collection Time
Time
Time
Time
Difference
Difference
Difference
Station
7/14/2009
7/15/2009
7/17/2009
7/14/09
7/15/09
7/17/09
EGIA100M-UP
10:41
9:41
12:00
EGIA30M-DN
14:27
10:43
13:20
3:46
1:02
1:20
EGIA100M-DN
16:30
11:45
14:30
2:03
1:02
1:10
EGIA300M-DN1
15:25
12:47
15:42
1:05
1:02
1:12
Notes:
1 - On July 14, the sample from EGIA300M-DN was collected before that obtained from EGIA100M-DN.
Hudson River PCBs Site
EPA Phase I Evaluation Report - Addendum
Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Table l-B-2
Time of Travel for Near-Field PCB Release Mechanism
East Griffin Island Area
Average of
Average of
Average of
Time of
Time of
Velocity (m/s)
Velocity (m/s)
Velocity (m/s)
Time of Travel
Travel
Travel
Location
Distance (m)
7/14
7/15
7/17
(min)7/14
(min)7/15
(min)7/17
100 m up
0.164
0.163
0.126
30 m down
350
0.194
0.193
0.151
35.57
35.79
46.30
100 m down
100
0.187
0.186
0.145
8.59
8.64
11.04
300 m down
133
0.148
0.147
0.114
11.86
11.93
15.26
Hudson River PCBs Site
EPA Phase I Evaluation Report - Addendum
Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Dissolved PCB
Particulate PCB
TSS
Date
Location
{ng/U
(ngft)
(rng/l)
7/14/2009
EGIA100M-UP
248,0
10,9
1,1
7/15/2009
EGIA100M-UP
448,2
16,9
1,9
7/17/2009
EGIA1G0M-UP
839.0
14,4
1,3
Glens Falls
Fort Edward
Lock #7
T.I. Dam
S Lock #6
Lock #5
Schuylerville
Stillwater Jh
Dissolved PCB
Particulate PCB
TSS
Date
Location
(ng/L)
(ng/L}
(mg/Lj
7/14/2009
E&ABQtMH
1,558-4
182.4
5.0
7/15/2009
£GIA3QM-m
1,070.6
81.3
1.8
7/17/2009
EGiA30fVl-DIM
973.0
32.8
2.1
Date
Location
Dissotved PCB
(tig/D
Participate PCB
ing/l-J
TSS
{ms/U
7/14/2009
EGIA100M-DN
253.5
S7.S
1.7
7/15/2009
EGIA100M-DN
964.5
52.9
2.4
7/17/2009
EGIA100M-DN
3,970.0
275.1
2.7
Dale
locaHofi
Dissolved PCS
{fgA)
Partial iale PC0
{"gA)
TSS
ftng/t)
7/14/2009
E&SA300M-DN
813.3
95,4
2.4
7/15/2009
E6SA300M-DN
1,355-3
98.3
2.8
7/17/2009
ESiA300M-DN
8,220.0
22ft2
6,9
o
Notes:
CU IDs labetedlin black.
Source: Anchor OEA ,2009.
Phase 1 Data Compilation.
Hudson River PCBs Site.
Figure 2.8-1. November 2009.
LEGEND
Sampling Transect
Upstream Sampling Point
Champlain Canal
~ Certification Units
Dams and Locks
Shore Line
—— River Miles
Near- Field PCB Release Mechanism Study Results
EPA Phase 1 Evaluation Report -Addendum - Hudson River PCBs Site
Figure 1-B-l
April 2010
-------
TOPIC 1-C
NEAR-FIELD PCB TRANSECT STUDY
-------
Topic 1-C - Near-Field PCB Transect Study
During Phase 1 dredging, GE collected samples along transects in the near-field. Two types of
near-field transects were collected. The first, which was associated with the near-field daily
monitoring program, is turbidity/TSS transect studies, in which turbidity was monitored along
transects located upstream, adjacent to, and downstream of dredging operations. TSS was
measured at the point of maximum turbidity within each transect. The second type of transect
studies, called PCB transect studies, consisted of PCB monitoring along transects perpendicular
to river flow during May, July, August, and September 2009 as described in the Phase 1 Data
Compilation report (GE 2009). While these transects represent instantaneous observations in a
very dynamic near-field area, and mostly did not follow a time of travel sampling approach, GE
has used the PCB transects results to determine load gains and losses, compared them to 24-hour
composite loads calculated in the far-field, and made inferences about the redistribution of PCBs
downstream of dredging. Based on GE's analysis (GE, 2010) they concluded that:
• PCBs were lost between the Rogers Island and EGIA dredging areas and the Thompson
Island Dam far-field station, with average declines in concentration of 0.5 kg/day
between Rogers Island and the Thompson Island Dam and 1.4 kg/day between the EGIA
and the Thompson Island Dam, for a total of 1.9 kg of PCBs lost per day.
• The decline lessened between Lock 7 and just upstream of CU-17 over the course of the
transect studies, possibly reflecting a growing contribution to the river from re-suspended
sediments which had deposited in the Thompson Island Pool. The impact of re-
distributed sediments is also demonstrated in the increase in PCB loads through the
Thompson Island Pool measured during the August 10, 2009 sampling event, conducted
while there was no dredging. The percentage of PCBs released by dredging operations
that did not reach the Thompson Island station could actually be higher because losses
may have been partially masked by PCB releases from re-distributed sediments (GE,
2010).
Because the data collected and used to develop the above conclusions regarding load was
collected in the near field and is thus highly dependent on several different factors at any time
during the day, it is viewed by the EPA as instantaneous. Collection of an instantaneous sample
within a plume is dependent of conditions upstream at the dredge head, including time of travel,
bucket decant, presence of NAPL, and the closure or non-closure of the dredge bucket due to
debris. These conditions vary in the near field suggesting that the transect results are subject to
high variability and uncertainty. The objective of this section is to present evidence that supports
the instantaneous and highly variable nature of these transects and while useful, they should be
used with considerable caution when generalized over the dredging period.
1-C.l. Comparison of Transect and Buoy TSS Measurements
In order to determine if instantaneous data could be used to give a reasonable picture of daily
load, TSS concentrations obtained from samples collected in the morning were compared to
those measured in samples collected during the afternoon during the twice daily turbidity
transect observations in the near-field. Samples obtained from the Lock 7 and West Rogers
Island transects identified below were used in the analysis:
Hudson River PCBs Site Page 1-C-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
• 100 m upstream
• 10m side channel
• 100 m downstream
• 300m downstream
One set of paired data from each transect was analyzed from Lock 7, and four sets from each
transect were analyzed from the West Channel of Rogers Island. The comparison, shown on
Figures 1-C-la through 1-C-le, shows that the morning and afternoon concentrations are not
consistent, indicating that there is too much variability inherent in these measurements in order
for them to be used to represent load for an entire day.
This conclusion is further supported by comparison of transect and buoy data from Lock 7 (100
m upstream and 300 m downstream), the West Channel of Rogers Island (100 m upstream and
300 m downstream), and the East Channel of Rogers Island (downstream). The TSS
concentrations measured in the transect samples and buoy samples were also dissimilar; in most
cases, the transect TSS values exceeded those observed in the buoy samples, indicating that the
transect values may overestimate the amount of suspended solids in the water column.
Comparison of transect and buoy results is shown on Figures l-C-2a through l-C-2f.
1-C.2. TSS Loading
Examination of TSS loads calculated using transect and buoy results for both the Rogers Island
area and the EGIA further support the conclusion that the near-field transects data in the water
column are highly variable. Comparisons of these loads are presented on Figures l-C-3 and 1-C-
4, respectively. In the Rogers Island area, the TSS loads yielded by the transect data and buoy
data are similar during the time period of May 2009 through the beginning of October 2009, with
the exception of several instances of elevated load in the middle to the end of August 2009 based
on the buoy TSS results. In approximately the middle of October 2009, the TSS loads diverge,
with the transect loads exceeding the buoy loads in the majority of instances (Figure l-C-3).
In the EGIA, TSS loads were calculated for transect and buoy data on the following dates:
• July 23, 2009
• August 12, 2009
• August 14, 2009
• August 17, 2009
• August 22, 2009
Comparison of transect and buoy TSS loads for these dates indicates that for all dates but August
14 and August 22, 2009, the loads obtained using transect data exceeded those developed with
the buoy data. In the cases of July 23 and August 17, 2009, the discrepancy was more significant
for either the upstream or downstream load comparison, or for both.
On July 23, 2009, the upstream transect loads were approximately 40 tons/day, a value 5 times
greater than the upstream buoy load of approximately 8 tons/day. The downstream transect load,
approximately 40 tons/day, was 5 times greater than the downstream east and west buoys loads,
also approximately 8 tons/day, and approximately 3 times greater than that (approximately 12
Hudson River PCBs Site Page l-C-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
tons/day) calculated based on the 100 m downstream buoy data. On August 17, 2009, the TSS
load obtained from the upstream transect data was approximately 100 tons/day, a value
approximately 7 times greater than the 15 tons/day calculated based on the upstream buoy. The
discrepancies between the loads based on the downstream transect data and those developed
using the downstream east and downstream west buoys were less pronounced, exceeding by only
1 to 2 times. The TSS load calculated using data from the 100 m downstream buoy was of a
similar magnitude. Comparisons of loads for all dates are shown on Figure l-C-4. During these
transect studies, the propeller wash of the sampling boat induced resuspension may have
compromised the evaluation of TSS load changes between Rogers Island and East Griffin Island.
In Section 5.1.1.2 of the GE Phase 1 Evaluation Report (GE, 2010), GE also attributes the
increases in TSS to sediment resuspension caused by operation of the sampling vessel in very
shallow water.
1-C.3. Instantaneous PCB Loading
Analysis of PCB load data developed using transect study data indicated that loads were not
uniform across transects. For example, the loads calculated for each node sampled along the
WRI transect on May 26, 2009 ranged from 14 to 651 g/day, with the smallest load observed at
Node # 8 (see Figure 1-4-2 in EPA's Phase 1 Evaluation Report [EPA, 2010] and Figure 2.8-15
in GE's Phase 1 Data Compilation Report [GE, 2009]), closest to the west bank of Rogers Island
(also the eastern most node). The loads calculated along the ERI transect on that date ranged
from 6 to 17 g/day, with the smallest load observed at Node #4, one node west of that closest to
the east bank of the river. While there was no pattern of consistent increase or decline in either
direction along the WRI and ERI transects, in both instances, the highest loads were present at
the westernmost and central nodes. The loads calculated along the NTIP transect on that date
ranged from 24 to 149 g/day, with the smallest load observed at Node # 1, located closest to the
east bank of the river. In this case, the highest loads were present at the central nodes and eastern
nodes. The variability in concentrations between transect nodes suggests that equal volume
compositing, done at some stations during the May 2009 transect studies, are likely biased. In
Section 5.1.1.2 of the GE Phase 1 Evaluation Report (GE, 2010), GE has also acknowledged that
the transect results may be biased high due to the sampling technique, which included using
equal volumes from five nodes across the river to form a single composite sample when the
hydrologic data collected in May 2009 from this area showed that the westernmost node
represents only one percent of the total flow of the river. This uncertainty strongly indicates that
the transect data must be interpreted with due caution as the lack of mixing along the cross
section might be obscured if equal aliquots are taken to form a sample as was done for the NTIP
transect.
Loads calculated along the EGIA-Down transect on July 23, 2009 ranged from 214 to 2406
g/day, with the smallest load observed at Node # 1, closest to the west bank of the river. In this
instance, the highest loads were observed in the central nodes. During the July 2009 transect
studies, because of variability in flow, and uncertainties related to differences in time of travel,
sample compositing, the analytical measurements and presence of NAPL, instantaneous load
differences of about 12 mg/second cannot be discerned as being significant (see Section 1-4.2.1
in EPA's Phase 1 Evaluation Report (EPA, 2010).
Transect loads are presented in Table 1-C-l.
Hudson River PCBs Site Page l-C-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
1-C.4. Daily PCB Loading
Daily PCB loads were calculated based on instantaneous transect PCB data for the following
stations:
• North WRI
• WRI
• North ERI
• ERI
• NTIP
• EGIA-Up
• EGIA-Curtain
• EGIA-Down
Analysis of the loads observed at these stations (not including the EGIA-Curtain, as it is located
inside the silt curtain and may not be representative of river conditions) indicated a general
increase in PCB loading in the downstream direction. Examination of daily loads on the dates
where transect sampling was performed in both the Rogers Island Area and EGIA (shown on
Figure l-C-5) indicated that maximum daily PCB loads were generally observed at the NTIP
transect, with approximate values of 1500 g/day on 8/10/09, 4000 g/day on 8/14/09, and 2600
g/day at 8/17/09, and at the EGIA-Down transect, with approximate values of 1300 g/day on
8/12/09, 3000 g/day on 8/22/09, and 2800 g/day on 10/21/09. During the August 2009 transect
studies, the propeller wash of the sampling boat induced resuspension which may have
compromised the evaluation of PCB load changes between Rogers Island and East Griffin Island
since this may be a factor even when TSS differences do not appear to be large. In Section
5.1.1.2 of the GE Phase 1 Evaluation Report (GE, 2010), GE also attributes the increases in TSS
to sediment resuspension caused by operation of the sampling vessel in very shallow water. In
the 8/14/09 sampling event, a load loss occurred between the EGI-Up and EGI-Down transects,
suggesting that uncertainties in these data and the daily variability in near-field concentrations
preclude an assessment of whether the concentration and load differences are statistically
significant. On 7/23/09, the maximum load of approximately 9,000 g/day was observed at the
EGIA-Up transect. The corresponding load at the EGIA-Down transect for that date was
approximately 7,500 g/day. Another point worth noting is that for the July 2009 transect studies,
the sampling event conducted by GE started at the EGIA-Down transect and proceeded towards
the EGIA-Up transect. The time of travel approach was not followed.
Though there are no calculations of PCB load based on buoy results to compare to those obtained
using the transect results, the conclusion stated above for TSS loads applies here, as well. The
inherent variability in the near-field precludes the use of instantaneous measurements to evaluate
PCB load added to the water column due to dredging.
1-C.5. Summary
The above analysis strongly suggests that the near-field area is highly dynamic and instantaneous
measurements should be treated as such because they represent a snap shot at the time of
collection. While the PCB transects do shed light on the nature of the near-field PCBs, in terms
of dissolved and particulate phases, uncertainties related to this data indicate caution in
generalizing their meaning. GE relies on instantaneous measurements that are representative of a
Hudson River PCBs Site Page l-C-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
snapshot to calculate daily PCB and TSS loads and then bases further analysis for the entire
dredging period on such calculations. Such conclusions are not valid for the intended purpose of
the data.
References
GE, 2009. Phase 1 Data Compilation Report. Prepared for General Electric Company, Albany,
NY, by Anchor QEA, LLC. November 2009.
GE, 2010. Phase 1 Evaluation Report. Prepared for General Electric Company, Albany, NY by
Anchor QEA, LLC and Arcadis. March 2010.
Hudson River PCBs Site Page l-C-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 1-C-l
PCB Transect Study Loading Analysis
Location111
Date
Total PCB
(ng/Lf1
Diss. PCB
(ngAf
% Diss.
PCB m
Part. PCB
(mg/kg)
TSS
(mg/L)
Mean USGS Flow
at Fort Edward
During Sampling
(cfs)
% Flow per
Transect
Node
Flow
Factor
Node or
Transect
Flow (cfs)
PCB
Loading
(g/day)
WRI-1
5/22/2009
853.0
487.0
57.1
26.0
14.1
8993
3.9
0,96
334
696
WRI-2
5/22/2009
1284,0
310,9
24,2
31.4
31.0
8993
8.1
0.96
704
2212
WRI-3
5/22/2009
541,9
121.7
22,5
30.7
13.7
8993
9.8
0,96
850
1126
WRI-4
5/22/2009
47.0
25.5
543
3.4
6,4
8993
17.5
0.96
1516
174
WRI-5
5/22/2009
32.4
11.1
34.3
5.1
4.2
8993
21.6
0.96
1871
148
WRI-6
5/22/2009
6.9
19.1
100.0
.J4J
4.2
8993
15.6
0.96
1348
23
WRI-7
5/22/2009
11.8
15.8
100.0
J4)
4,9
8993
15.5
0,96
1338
39
WRI-8
5/22/2009
3.2
20.9
100.0
J4)
3,8
8993
8.0
0.96
696
5
ERI-1
5/22/2009
189.6
154,0
81.2
9.5
3.7
9340
28.0
0.04
98
45
ERI-2
5/22/2009
202.4
151.0
74,6
18.9
2.7
9340
22.1
0,04
7?
38
ERI-3
5/22/2009
199,8
146.0
73,1
17.2
3,1
9340
26.5
0,04
93
45
ERI-4
5/22/2009
192.0
167,2
87.1
8.1
3.1
9340
9.8
0,04
34
16
ERI-5
5/22/2009
242,5
161,0
66.4
19.6
4,2
9340
13.6
0.04
48
28
NTIP-1
5/22/2009
177.2
161,0
90.9
3.9
4.1
8890
1.1
1.0
97
42
NTIP-2
5/22/2009
160.9
87.8
54.6
21,2
3,5
8890
6.4
1.0
572
225
NTIP-3
5/22/2009
79.9
74.1
92.7
1,4
4.1
8890
22,6
1.0
2010
393
NTIP-4
5/22/2009
35.0
45.7
100,0
3.6
8890
403
1.0
3579
306
NTIP-S
5/22/2009
54.2
24.2
44,6
8.0
3.8
8890
29.6
1.0
2631
348
WRI-1
5/25/2009
92.2
NA
--
--
13
6900
3.9
0.96
256
58
WRI-2
5/25/2009
54.9
NA
1.4
6900
8.1
0,96
540
73
WRI-3
5/25/2009
3.0
NA
1.2
6900
9.8
0.96
652
5
WRI-4
5/25/2009
19.0
NA
..
1.2
6900
17.5
0.96
1163
54
WRI-5
5/25/2009
14.7
NA
1.2
6900
21.6
0,96
1435
52
WRI-6
5/25/2009
3.0
NA
..
..
1.4
6900
15,6
0.96
1035
8
WRI-7
5/25/2009
3.2
NA
1,2
6900
15.5
0.96
1026
8
WRI-S
5/25/2009
15.8
NA
..
1.7
6900
8.0
0.96
534
21
ERI-1
5/25/2009
533
NA
1.1
7647
28,0
0,04
80
10
ERI-2
5/25/2009
71.3
NA
--
1.2
7647
22.1
0,04
63
11
ERI-3
5/25/2009
79.0
NA
1.4
7647
26.5
0,04
76
15
ERI-4
5/25/2009
74.9
NA
„
2,7
7647
9.8
0,04
28
5
ERI-5
5/25/2009
66.8
NA
..
..
1.5
7647
13.6
0.04
39
6
NTIP-1
5/25/2009
57.4
NA
2.0
6750
1.1
1.0
74
10
NTIP-2
5/25/2009
44.6
NA
-
13
6750
6.4
1.0
434
47
NTIP-3
5/25/2009
35,1
NA
..
1.4
6750
22.6
1.0
1526
131
NTIP-4
5/25/2009
16.8
NA
..
1.6
6750
403
1.0
2718
112
NTIP-5
5/25/2009
8.2
NA
1.5
6750
29,6
1.0
1998
40
WRI-1
5/26/2009
429.7
NA
7,2
8276
3.9
0,96
307
322
WRI-2
5/26/2009
410.4
NA
--
--
8,9
8276
8.1
0.96
648
651
WRI-3
5/26/2009
221.4
NA
5,8
8276
9.8
0,96
782
423
WRI-4
5/26/2009
155.6
NA
--
4,1
8276
17,5
0.96
1395
531
WRI-5
5/26/2009
36.5
NA
3.0
8276
21.6
0.96
1721
154
WRI-6
5/26/2009
69.1
NA
-
4,1
8276
15.6
0,96
1241
210
WRI-7
5/26/2009
18.8
NA
..
3.2
8276
15,5
0,96
1231
57
WRI-8
5/26/2009
9.1
NA
3.6
8276
8.0
0,96
641
14
ERI-1
5/26/2009
94.7
NA
-
1.2
6849
28,0
0.04
72
17
ERI-2
5/26/2009
94.1
NA
..
1.2
6849
22,1
0.04
57
13
ERI-3
5/26/2009
88.8
NA
..
1.3
6849
26,5
0,04
68
15
ERI-4
5/26/2009
1003
NA
..
1.5
6849
9.8
0.04
25
6
ERI-5
5/26/2009
96.0
NA
1.2
6849
13.6
0.04
35
8
NTIP-1
5/26/2009
133,8
NA
2.2
6753
1.1
1.0
74
24
NTIP-2
5/26/2009
75.2
NA
2,0
6753
6.4
1.0
435
80
NTIP-3
5/26/2009
39.8
NA
1.6
6753
22.6
1.0
1527
149
NTIP-4
5/26/2009
16.0
NA
..
1,7
6753
403
1.0
2719
106
NTIP-5
5/26/2009
29.4
NA
-
1,9
6753
29.6
1.0
1999
144
WRI-1
5/28/2009
1620,0
NA
..
..
73
5975
3.9
0,96
222
878
WRI-2
5/28/2009
1542.8
NA
..
7.9
5975
8.1
0,96
468
1766
WRI-3
5/28/2009
494.5
NA
4.0
5975
9.8
0.96
565
683
WRI-4
5/28/2009
192.4
NA
--
2,3
5975
17.5
0.96
1007
474
WRI-5
5/28/2009
83.9
NA
..
1,8
5975
21.6
0,96
1243
255
Hudson River PCBs Site Page 1 of 3
EPA Phase I Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 1-C-l
PCB Transect Study Loading Analysis
Location 111
Date
Total PCB
(ng/Lf1
Diss. PCB
(ng/L)'21
% Diss.
PCB 131
Part. PCB
(mg/kg)
TSS
(mg/L)
at Fort Edward
During Sampling
(cfs)
% Flow per
Transect
Node
Flow
Factor
Node or
Transect
Flow (cfs)
PCB
Loading
(g/day)
WRI-6
5/28/2009
11.8
NA
1.6
5975
15.6
0.96
896
26
WRI-7
5/28/2009
3.0
NA
--
1.8
5975
15.5
0.96
889
7
WRI-8
5/28/2009
9.1
NA
1.8
5975
8.0
0.96
462
10
ERI-1
5/28/2009
3000.0
NA
9.6
3860
28.0
0.04
40
296
ERI-2
5/28/2009
3070.0
NA
8.2
3860
22.1
0.04
32
240
ERI-3
5/28/2009
5098.0
NA
8.5
3860
26.5
0.04
38
477
ERI-4
5/28/2009
4127.0
NA
7.8
3860
9.8
0.04
14
143
1 ERI-5
5/28/2009
4920.0
NA
7.9
3860
13.6
0.04
20
237
NTIP-1
5/28/2009
360.0
NA
--
2.1
7478
1.1
1.0
82
72
NTIP-2
5/28/2009
361.0
NA
~
2.5
7478
6.4
1.0
481
425
NTIP-3
5/28/2009
431.9
NA
2.9
7478
22.6
1.0
1691
1785
NTIP-4
5/28/2009
204.8
NA
2.1
7478
40.3
1.0
3011
1508
NTIP-5
5/28/2009
230.9
NA
2.5
7478
29.6
1.0
2213
1250
WRI-1
7/23/2009
1208.1
777.2
64.3
93.1
4.6
6085
20
0.96
1171
3460
WRI-2
7/23/2009
229.0
214.5
93.6
7.4
2.0
6085
20
0.96
1171
656
WRI-3
7/23/2009
127.2
55.4
43.5
21.3
3.4
6085
20
0.96
1171
364
| WRI-4
7/23/2009
384.5
238.5
62.0
60.6
2.4
6085
20
0.96
1171
1101
WRI-5
7/23/2009
369.6
317.7
86.0
38.1
1.4
6085
20
0.96
1171
1059
ERI-Comp.
7/23/2009
6678.2
5141.7
77.0
113.0
13.6
6070
100
0.04
227
3707
NTIP-Comp.
7/23/2009
590.9
548.0
92.7
16.3
2.6
5930
100
1.0
5930
8568
EGIA-Up
7/23/2009
623.1
523.9
84.1
31.3
3.2
5930
100
1.0
5930
9034
EGIA - Inside Sheeting
7/23/2009
21687.5
9464.7
43.6
745.3
16.4
EGIA-DS of Sheeting -1
7/23/2009
611.1
545.8
89.3
18.3
3.6
5953
20
1.0
1191
1779
EGIA-DS of Sheeting - 2
7/23/2009
623.7
481.8
77.2
44.9
3.2
5953
20
1.0
1191
1816
EGIA-DS of Sheeting - 3
7/23/2009
682.1
551.4
80.8
55.1
2.4
5953
20
1.0
1191
1986
EGIA-DS of Sheeting - 4
7/23/2009
672.7
587.4
87.3
50.2
1.7
5953
20
1.0
1191
1958
EGIA-DS of Sheeting - 5
7/23/2009
928.7
651.9
70.2
63.8
4.3
5953
20
1.0
1191
2703
EGIA-Down-1
7/23/2009
564.2
518.6
91.9
14.2
3.2
5176
3
1.0
155
214
EGIA-Down-2
7/23/2009
592.3
459.0
77.5
41.8
3.2
5176
26
1.0
1346
1949
EGIA-Down-3
7/23/2009
528.1
461.8
87.4
19.1
3.5
5176
36
1.0
1863
2406
EGIA-Down-4
7/23/2009
648.7
535.1
82.5
31.5
3.6
5176
27
1.0
1398
2217
EGIA-Down-5
7/23/2009
752.7
705.2
93.7
25.6
1.9
5176
8
1.0
414
762
Tl Stilling Well Grab
7/23/2009
366.2
320.4
87.5
17.6
2.6
-
WRI-North
8/10/2009
5.6
4.6
81.3
0.8
1.4
7269.0
100
0.96
6997
96
WRI-Comp.
8/10/2009
33.0
28.4
86.1
2.6
1.8
7269.0
100
0.96
6997
564
ERI-North
8/11/2009
48.7
24.6
50.4
9.4
2.6
7269.0
100
0.04
272
32
ERI-Comp.
8/10/2009
186.9
127.0
68.0
3.1
19.2
7269.0
100
0.04
272
124
NTIP-Comp.
8/10/2009
82.0
25.6
31.3
9.8
5.7
7269.0
100
1.0
7269
1457
EGIA-Up
8/10/2009
48.4
31.6
65.3
6.0
2.8
7269.0
100
1.0
7269
860
EGIA - Inside Sheeting
8/10/2009
21692.1
14773.2
68.1
1438.4
4.8
7269.0
--
EGIA-DS of Sheeting
8/10/2009
71.5
56.2
78.6
6.0
2.6
7269.0
--
EGIA - Inside Silt Curtain
8/10/2009
86.8
58.2
67.0
14.5
2.0
7269.0
EG 1 A-Down-1
8/10/2009
44.0
38.4
87.4
2.9
1.9
7269.0
3
1.0
218
23
EGIA-Down-2
8/10/2009
43.1
31.0
72.0
5.2
2.3
7269.0
26
1.0
1890
199
EGIA-Down-3
8/10/2009
57.2
33.9
59.3
CO
CO
2.6
7269.0
36
1.0
2617
366
EGIA-Down-4
8/10/2009
62.7
41.5
66.2
8.7
2.4
7269.0
27
1.0
1963
301
EGIA-Down-5
8/10/2009
112.5
78.9
70.2
18.6
1.8
7269.0
8
1.0
582
160
WRI-North
8/12/2009
18.9
CO
CO
46.4
3.9
2.6
6767.0
100
0.96
6514
300
WRI-Comp.
8/12/2009
64.7
73.5
100.0
.J4'
3.1
6767.0
100
0.96
6514
1031
ERI-North
8/12/2009
234.3
85.1
36.3
14.9
10.0
6767.0
100
0.04
253
145
ERI-Comp.
8/12/2009
817.8
533.3
65.2
54.9
5.2
6767.0
100
0.04
253
506
NTIP-Comp.
8/12/2009
59.8
49.1
82.0
2.5
4.3
6767.0
100
1.0
6767
990
EGIA-Up
8/12/2009
74.9
73.9
98.6
0.5
2.3
6767.0
100
1.0
6767
1240
EGIA - Inside Sheeting
8/12/2009
26254.2
17210.0
65.6
611.1
14.8
6767.0
EG 1 A-Down-1
8/12/2009
77.1
64.0
83.1
5.0
2.6
6767.0
3
1.0
203
38
EGIA-Down-2
8/12/2009
71.3
45.4
63.7
9.6
2.7
6767.0
26
1.0
1759
307
EGIA-Down-3
8/12/2009
66.8
53.8
80.6
4.8
2.7
6767.0
36
1.0
2436
398
EGIA-Down-4
8/12/2009
85.3
79.6
93.4
2.7
2.1
6767.0
27
1.0
1827
381
EGIA-Down-5
8/12/2009
150.0
132.7
88.5
8.2
2.1
6767.0
8
1.0
541
199
Hudson River PCBs Site Page 2 of 3
EPA Phase I Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 1-C-l
PCB Transect Study Loading Analysis
Mean IM5 flow
at Fort Edward
% Flow per
Node or
PCB
Total PCB
Diss. PCS
% Diss.
Part. PCB
TSS
During Sampling
Transect
Flow
Transect
Loading
Location113
Date
(ng/tf1
PCB£3)
(mg/L)
(cfs)
Node
Factor
Flow (cfs)
(g/day)
WRI-North
8/14/2009
14.8
10.3
70.0
2,7
1.6
4044,0
100
0.96
3893
141
WRI-tomp.
8/14/2009
223.5
161.1
72.1
15,8
4.0
4044.0
100
0.96
3893
2127
ERI-North
8/14/2009
933
35.7
38.3
8,0
7.2
4044.0
100
0.04
151
35
ERI-Gomp.
8/14/2009
3993.5
2277.1
57.0
195.9
8.8
4044,0
100
0.04
151
1477
NTIP-Comp.
8/14/2009
415.8
248.6
59.8
43
38,8
4044.0
100
1.0
4044
4112
EG lA-Up
8/14/2009
206.8
126.9
61.3
28.6
2.8
4044,0
100
1.0
4044
2045
EGIA - Inside Sheeting
8/14/2009
32038.6
16592.8
51.8
141.7
109.0
4044,0
EGIA-Down-1
8/14/2009
137.1
105.9
77.3
14.4
2,2
4044.0
3
1.0
121
41
EGIA-Down-2
8/14/2009
163,0
123.1
75.5
20.1
2.0
4044,0
26
1.0
1051
419
EGIA-Down-3
8/14/2009
173,0
117.6
68,0
22.9
2.4
4044.0
36
1.0
1456
616
EGIA-Down-4
8/14/2009
168.0
122.6
73.0
20.9
2.2
4044.0
27
1.0
1092
448
EGIA-Down-5
8/14/2009
197.6
137.1
69.4
21.2
2.9
4044.0
8
1.0
324
156
WRI-North
8/17/2009
140.6
67.8
48.2
25.8
2,8
3711.0
100
0.96
3572
1228
WRI-Comp.
8/17/2009
155.8
97.6
62.6
213
2.7
3711.0
100
0.96
3572
1361
ERI-North
8/17/2009
68.5
25.1
OZ
20.1
3711.0
100
0.04
139
93
ERI-tomp,
8/17/2009
h nb
869.5
55.4
6.0
3711,0
100
0.04
139
532
NTIP-Comp.
8/17/2009
288.6
133.6
46.3
13
117.0
3711,0
100
1.0
3711
2618
EGIA-Up
8/17/2009
192.1
88.7
46.2
9.2
113
3711.0
100
1.0
3711
1743
EGIA - Inside Sheeting
8/17/2009
25444,1
25420.1
99.9
0.7
32.6
3711,0
EGIA-Down-1
8/17/2009
188.2
100.5
53,4
26.5
33
3711.0
3
1.0
111
51
EGIA-Down-2
8/17/2009
193.7
96.1
49.6
49.8
2.0
3711.0
26
1.0
965
457
EGIA-Down-3
8/17/2009
227.2
102.5
45.1
77.0
1.6
3711.0
36
1.0
1336
742
EGIA-Down-4
8/17/2009
237.2
119.7
50.5
653
1.8
3711.0
27
1.0
1002
581
EGIA-Down-5
8/17/2009
434.7
290.0
66.7
51.2
2.8
3711,0
8
1.0
297
316
WRI-North
8/22/2009
43.9
27.0
61.6
53
3,2
2381.0
100
0.96
2292
246
WRI-Comp.
8/22/2009
105.8
55.4
52.4
25.1
2.0
2381.0
100
0.96
2292
593
ERI-North
8/22/2009
SSb
37.1
12.9
4,4
57,7
2381.0
100
0.04
89
63
ERI-tomp.
8/22/2009
u
58.8
103.4
9.9
2381.0
100
0.04
89
540
NTIP-Comp.
8/22/2009
235.0
188.8
80.4
1.5
30.8
2381,0
100
1.0
2381
1368
EGIA-Up
8/22/2009
187.0
138.9
74.3
28.8
1,7
2381.0
100
1.0
2381
1088
EGIA - Inside Sheeting
8/22/2009
3473,0
2609.0
75,1
326.0
2.7
2381.0
"
--
EGIA - Inside Silt Curtain
8/22/2009
3102.6
2375.3
76.6
60.6
12,0
2381.0
EGIA-Down-1
8/22/2009
191.1
115.2
60,3
11.6
6.6
2381,0
3
1.0
71
33
EGIA-Down-2
8/22/2009
157.3
109.8
69.8
29.2
1.6
2381,0
26
1.0
619
238
EGIA-Down-3
8/22/2009
598.7
415.6
69.4
130.8
1.4
2381.0
36
1.0
857
1255
EGIA-Down-4
8/22/2009
623.7
477.9
76.6
92.9
1.6
2381.0
27
1.0
643
980
EGIA-Down-5
8/22/2009
1220.1
894.6
73.3
' 25.2
2.6
2381.0
8
1.0
190
568
WRI-Comp,
10/21/2009
41.0
MA
..
NA
4039.0
100
0.96
3888
390
ERI-tomp,
10/21/2009
688.3
NA
NA
4039.0
100
0.04
151
254
NTIP-Comp.
10/21/2009
70.9
NA
..
NA
4039,0
100
1,0
4039
700
EGIA-Up
10/21/2009
142.5
NA
NA
4039.0
100
1.0
4039
1407
EGIA-Down
10/21/2009
272.2
NA
-
NA
4039,0
100
1.0
4039
2688
Notes:
(1). Sample locations illustrated in Figure 5.1-2.
(2) Revised correction factor developed during RAMP has been applied to PCB data,
(3) For paired samples where dissolved fraction exceeds total fraction, % dissolved is assumed to be 100%.
(4) For paired samples where dissolved fraction exceeds total fraction particulate fraction not calculated
WRI = West Rogers Island
ERI = East Rogers Island
NTIP = Northern Thompson Island Pool
EGIA- East Griffin Island Area
Source: Anchor QEA and ARCADIS,2010. Phase 1 Evaluation Report. Hudson River PCBs Site. Table 5.1-1.
March 2010.
Hudson River PCBs Site Page 3 of 3
EPA Phase I Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
35
30
~ 25
J
"3d
M, 20
c
0
1 15
10
5
0
Lock 7 Operation #1: 100m Downstream Transect
5 10 15 20
Morning TSS (mg/L)
Lock 7 Operation #1: 10m Side Channel Transect
25
1:1 Line/'''
o
O
O o
o
o
10 15 20
Morning TSS (mg/L)
25
30
35
Lock 7 Operation #1: 100m Downstream Transect
10 15 20 25 30
Morning TSS (mg/L)
35
40 45
Lock 7 Operation #1: 300m Downstream Transect
10 15
Morning TSS (mg/L)
20
25
Near-field TSS Transect AM versus PM
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-C-la
April 2010
B
-------
West Rogers Island Operation #1: 100m Downstream Transect
E
30
25
20
15
10
1:1 Line
0
oo
o
10 15 20
Morning TSS (mg/L)
25
30
West Rogers Island Operation #1: 10m Side Channel Transect
1:1 Line.
~J
E
o
o
o
o
o O
0 --
2 3 4
Morning TSS (mg/L)
West Rogers Island Operation #1: 100m Upstream Transect
3.5
~J 2.5
OJD
S,
c ?
o ^
1.5
0.5
1:1 Line
O
0
0
o
°
° \
0 \
0 \
0 0.5 1
1.5 2 2.5
Morning TSS (mg/L)
3.5
West Rogers Island Operation #1: 300m Downstream Transect
70
60
3 50
ex)
g
§ 40
o
s
v
% 30
20
10
1:1 Line..--''
y''
O
0
#qO
O
o
0 10 20
30 40 50
Morning TSS (mg/L)
60 70
Near-field TSS Transect AM versus PM
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-C-lb
April 2010
\
B
-------
West Rogers Island Operation #2: 100m Downstream Transect
West Rogers Island Operation #2: 100m Upstream Transect
0 0.5 1
1.5 2 2.5 3 3.5
Morning TSS (mg/L)
4 4.5 5
West Rogers Island Operation #2: 10m Side Channel Transect
"©JO
5
2 3 4
Morning TSS (mg/L)
4
3.5
3
^ 2.5
ex)
g
- 9
© L
o
s
% L5
i
0.5
0
1:1 Line
0
O
0 0.5 1
1.5 2 2.5
Morning TSS (mg/L)
3.5
West Rogers Island Operation #2: 300m Downstream Transect
0 0.5 1 1.5 2 2.5
Morning TSS (mg/L)
3.5
Near-field TSS Transect AM versus PM
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-C-lc
April 2010
\
B
-------
West Rogers Island Operation #3: 100m Downstream Transect
70
60
50
3
"wi . _
s 40
a
o
o
E 30
Qi
<
20
10
1:1 Line,.-*''
o
o°
c
O
SXlA' oO O
o
o
o
cc
10 20 30 40
Morning TSS (mg/L)
50
60
West Rogers Island Operation #3: 10m Side Channel Transect
70
60
3 50
OJD
s
§ 40
0
s
v
1 30
20
10
0
1:1 Line
o
o
o
. .-<3
$p°o 9)
o
10 20 30 40 50
Morning TSS (mg/L)
60
70
70
West Rogers Island Operation #3: 100m Upstream Transect
E
10 15 20 25
Morning TSS (mg/L)
30
35
West Rogers Island Operation #3: 300m Downstream Transect
35
30
25
5*
g 20
s
o
o
E 15
£
<
10
10 15 20
Morning TSS (mg/L)
25
30
40
1:1 Line
0
0
° 0
O O .../ O
^ n Q.--CO o
^ (9
O
O
o
35
Near-field TSS Transect AM versus PM
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 1-C-ld
April 2010
-------
West Rogers Island Operation #6: 100m Upstream Transect
20 30 40
Morning TSS (mg/L)
West Rogers Island Operation #6: 10m Side Channel Transect
70
60
50
"wi . _
s 40
a
©
o
E 30
0)
-------
Lock 7 Operation #1: 100m Upstream
10
Buoy TSS(mg/L)
15
J
~Sio
es
•-
H
20
Lock 7 Operation #1: 300m Downstream
4 6
Buoy TSS(mg/L)
10
Near-field TSS Buoy versus Transect
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-C-2a
April 2010
B
-------
1000 -
Rogers Island East Channel Downstream
100
J
"3)
£,
-------
4
J
~Sio
a 3
es
•-
H
2
1
West Rogers Island Operation #1: 100m Upstream
X
1:1 Line /
X
o
/
V
V
\
\
\
o o
o
X
o
0
o
0
o /
Q>
o
X
X
o
,s
x°
/
2 3
Buoy TSS(mg/L)
West Rogers Island Operation #1: 300m Downstream
10
20 30
Buoy TSS(mg/L)
Near-field TSS Buoy versus Transect
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
40
50
Figure l-C-2c
April 2010
\
B
-------
West Rogers Island Operation #2: 100m Upstream
0.5
1 1.5
Buoy TSS (mg/L)
2.5
3.5
^2.5
J
"oil
I ^
-------
West Rogers Island Operation #3: 100m Upstream West Rogers Island Operation #3: 300m Downstream
B
Near-field TSS Buoy versus Transect
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-C-2e
April 2010
-------
4
J
~Sio
a 3
es
•-
H
2
1
West Rogers Island Operation #6: 100m Upstream
1:1 Line/''
o
/'
X
X
\
\
\
\
cP\
\
\
\
\
\
\
/
o
2 3
Buoy TSS(mg/L)
25
20
J
~ei)
S, 15
(Z3
(Z3
H
West Rogers Island Operation #6: 300m Downstream
1:1 Line /'
U
a
u
H
10
°o
o
°
cO°0
O oo
10 15
Buoy TSS (mg/L)
20
25
Near-field TSS Buoy versus Transect
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-C-2f
April 2010
B
-------
1000
100
>
ra
w
a>
c
c
o
o
B 10
0)
E
T3
ra
o
CO
CO
A Downstream East
O Downstream Transect
~ Downstream West
1
Q
0 o
0.1
5/20/2009
dfe °°o
o ^
O CD
O© O o
-------
120
100
>
re
I 80
£
£
O
Upstream
Upstream - Transect
100 m Downstream
Downstream East
Downstream West
Downstream Transect
7/23/2009
8/10/2009
8/12/2009 8/14/2009
Sample Date
8/17/2009
8/22/2009
East Griffin Island Area Near-field Down- and Upstream Buoy versus Transect TSS Load Analysis
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-C-4
April 2010
-------
10000
9000
8000
7000
m 6000
u
5000
I North WRI
I WRI
North ERI
I ERI
NTIP
EGIA-UP
EGIA-Down
£ 4000
3000
2000
1000
JMi
7/23/2009
8/10/2009
8/12/2009
8/14/2009
Sample Date
8/17/2009
8/22/2009
10/21/2009
Near-field Daily Total PCB Loading Transect Average
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-C-5
April 2010
-------
TOPIC 1-D
PCB FATE AND TRANSPORT IN THE
FAR-FIELD DURING DREDGING
-------
Topic 1-D - PCB Fate and Transport in the Far-field during Phase 1 Dredging
1-D.l. Introduction
In order to better understand the water column PCB loads during the 2009 Phase 1 dredging
event, EPA has conducted a more complete analysis of the fate and transport of PCBs and TSS
within the Thompson Island Pool and within the Upper Hudson River. This analysis is based on
an examination of the changes in the PCB homolog distribution and the behavior of the
particulates with movement downstream from the dredging operations in the near-field to the
first far-field station at Thompson Island and then to Waterford. It also makes the analyses
performed by EPA in related topics (Topics 1-A through 1-C, and 3) more complete.
1-D.2. PCB Fate and Transport between the Near-field and Thompson Island
Several studies were conducted during Phase 1 that provide information on the transport of PCBs
between the dredging operations (the near-field) and the first far-field station at Thompson
Island. While the data obtained at Thompson Island was specifically designed to examine PCB
transport and estimate PCB loads, the samples obtained in the near field were diagnostic in
nature and do not provide time-integrated observations. Nonetheless, the properties of the near-
field samples can provide some information on the nature of PCBs at the point of release and
how the PCB mixture changes downstream.
To examine PCB transport, the mixture of PCBs {i.e., the PCB homolog distribution) was
compiled for near-field and far-field samples. The average mass fraction for each PCB homolog
was calculated for all far-field samples collected at the Thompson Island far-field station
between May 2009 and October 2009. This Phase 1 average was then compared to the average
homolog mass fractions observed in the PCB near-field release mechanism study samples
(collected on July 14, 15, and 17, 2009) and PCB transect study samples (collected on July 23,
2009, August 10-12, 14, 17, and 22, 2009, and October 21, 2009) to evaluate any changes in the
homolog pattern between the dredging operation (the near-field) and the far-field. Mass fractions
were determined based on whole water PCB values; for the near-field release mechanism data,
this value was calculated as the sum of the dissolved and particulate results.
Analysis of the monochlorobiphenyl fraction indicated that the amount of this homolog in the
water column declined in the downstream direction. The highest average percentage was
observed in the near-field release sample (43 percent), followed by the transect samples (24
percent), and lowest amount was observed in the in the far-field samples (16 percent).
The dichlorobiphenyl fraction variation was not as pronounced; the percentage of far-field whole
water PCBs composed of dichlorobiphenyl was 39 percent, a value that is consistent with what
was observed in the near-field release study samples (also 39 percent). The transect study
samples, however, were composed of 45 percent dichlorobiphenyl.
Unlike the monochlorobiphenyl and dichlorobiphenyl fractions, the remaining average homolog
fractions (trichlorobiphenyl through decachlorobiphenyl) all increased in the downstream
direction. The distribution of PCBs present in the far-field samples contained a greater
percentage of Tri+ homologs than the near-field and transect study samples; average Tri+
percentages in the far-field, transect, and near-field samples are 45, 31, and 18 percent,
Hudson River PCBs Site Page 1-D-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
respectively. Note that the Tri+ PCB composition of the Total PCBs (TPCB) observed at
Thompson Island also increases at higher river flows.
Average homolog mass fractions for the far-field TID station, the PCB transect study, and the
PCB near-field release mechanism study samples are shown in Figure 1-D-la. Figure 1-D-lb is a
ternary diagram of the mass fractions of mono, di and Tri+ homologs for the entire set of
samples examined here. In this diagram, the homolog mass fractions clearly show the differences
in mass fraction among the three sample groups. The gradual shift to the heavier homolog
spectrum from the near-field to transect to the far-field samples is evident in both
representations.
Two hypotheses may be advanced to explain the findings described above: either a loss of lower
chlorinated PCBs or a gain of Tri+ PCBs between the near-field and the far-field TID station. GE
has advocated the second hypothesis, i.e., that a gain in Tri+ PCBs is responsible for the shift,
and that re-distributed sediments are the source of the additional Tri+ PCBs observed in the far-
field. To accomplish this gain in Tri+ PCBs, an increase in suspended matter would need to
occur at the same time to deliver the Tri+ PCB to the water column. However, this hypothesis is
not supported by the observed TSS concentrations in the near field. TSS concentrations in the
buoys downstream of East Griffin Island Area (EGIA) were higher and more variable,
particularly later in the program, than TSS concentrations measured at Thompson Island. If
additional solids were re-suspended downstream of EGIA, the observed TSS concentrations at
Thompson Island would have been significantly higher than values at EGIA. Therefore, the first
hypothesis, i.e., a loss of lower chlorinated PCBs, is the likelier explanation for the shift in
homologs. It is likely that during transport from the near-field, the lighter, more volatile PCB
homologs are lost by gas exchange causing the shift towards higher Tri+ PCB fraction at
Thompson Island. Notably loss of PCBs to the air was documented on a number of occasions
during Phase 1, particularly when oil-phase was visible and concentrated, such as at CU-3 or in
the sheet pile area at CU-18. While gas exchange losses were most apparent in these focused
areas, the elevated local concentrations near the dredge operations likely create similar
opportunities for gas exchange. This process is consistent with the entire set of available data.
1-D.3. PCB Fate and Transport through the Upper Hudson River
This section examines the mass of TPCB and PCB homologs transported to the Thompson
Island, Schuylerville (Lock 5) and Waterford far-field stations during the active dredging period
(May 15, 2009 to October 27, 2009), as well as the PCB homolog patterns at these three stations
that existed under baseline conditions (i.e., prior to dredging between 2004 through 2008).
Associated TSS loads for the three far-field stations during dredging and under baseline
conditions were also evaluated. In addition, potential causes and/or mechanisms to which the
observed PCB and TSS loss/gain rates and patterns might be attributed are discussed.
1-D.4. PCB and TSS Mass Balance Calculations
The mass of PCB homologs at the three far-field stations and the PCB mass lost or gained
between Thompson Island and Lock 5 and between Lock 5 and Waterford were calculated for
both the baseline condition and the active dredging period. TSS loads at the three far-field
stations were also calculated.
Hudson River PCBs Site Page l-D-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
1 -D. 4.1. Mass Balance under Baseline Conditions
The average PCB homolog and TSS daily loads under baseline conditions for the period of May
to November (which corresponds to the active Phase 1 dredging period1) were compiled for the
years 2004 to 2008 using available BMP data and the AutoBeale Load Estimation Program
(Richards, 1998). AutoBeale which is a computer implementation of the Beale Ratio Estimator,
iteratively seeks out the discharge stratification and minimizes the variance of the load estimate
for a given set of data. The cumulative PCB and TSS masses at the three stations were then
estimated by summing the average daily loads for this time period. These results are shown in
Table 1-D-l and Figures l-D-2a and l-D-2b.
1 -D. 4.2. Mass Balance under Active Dredging
In a similar manner, the mass of TPCB and PCB homologs at the three far-field stations during
active dredging were calculated for the period from July 5 to October 27, 2009. The PCB
sampling results prior to July 5, 2009 were not included because congener data were not
available for the Thompson Island station prior to that date, and congener data are needed to
assess the PCB homolog pattern being transported downstream. Additionally, the AutoBeale
Estimator was not need for this calculation since the water column was sampled daily on a 24
hour basis during this period.
For the dredging period, the mass flux of each PCB homolog was calculated using the water
column concentration multiplied by the flow in the following equation:
- % * Fff * K x 7^- (Eq. 1-D-l)
1 rQS- <* ! *! ;ay ni" lVT K5
Where: MFpCB = Mass flux of PCB (gram/day)
Cw = PCB concentration in water column (ng/L)
Daily flow at the far-field station (m3/s)
Thompson Island flow = 1.03 X Ft. Edward flow2
Schuylerville flow = 1.07 X Ft. Edward flow2
Ft. Edward and Waterford flows were obtained from USGS
The cumulative PCB mass for each homolog was then calculated by adding the daily mass
fluxes. For example, the cumulative PCB mass for the period of July 5 through October 27,
2009 was calculated as follows:
(?c
Mir (Ea. i-P-i}
k "/wlij'iil
1 The period of May to November is used as a general baseline condition for comparison to the transport during the
dredging period. The minor difference in the time periods to be compared (May 1 to November 30 vs. May 15 to Oct
27 should not have any bearing on the comparison of general PCB transport behavior. Note that all in-river activities
related to the Phase 1 remediation actually ended on December 4, 2009. Demobilization of dock structures and
related floats continued through December 23, 2009.
2 These factors adjust for the increase in drainage area from Ft Edward to these monitoring locations. They are the
ratio of the drainage area at the station to the drainage area at Ft Edward as given in USGS Open-File Report 81-
1055 Wagner (1982).
Hudson River PCBs Site Page l-D-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Where MPCb = Mass of PCB (grams)
MFpcs = Mass flux of PCB (gram/day)
Similar calculations were used to obtain associated TSS loads. The results are shown in Table
l-D-2, and Figures l-D-3a and l-D-3b. This table and these figures also show PCB mass
fraction, mass loss/gain between stations and percent mass loss/gain. For PCBs, the percent
mass loss or gain is calculated as:
1-D.5. Evaluation of PCB and TSS Loads
The results of the PCB homolog and TSS load calculations under baseline conditions and during
active dredging are provided below.
Table 1-D-l shows baseline PCB homolog and TSS at the three far-field stations for the years
2004 to 2008. Table 1-D-l and Figure l-D-2a show the average of the 5-year period for each
homolog. The table and figure indicate a consistent mass increase from the Thompson Island
Station to Schuylerville and from Schuylerville to Waterford for all homologs, with the
exception of monochlorobiphenyls which increased only slightly between Thompson Island and
Schuylerville and then decreased slightly from Schuylerville to Waterford. Table 1-D-l and
Figure l-D-2b show the average TSS for the 5-year period at each station. The table and figure
indicate a consistent increase in average TSS load from Thompson Island Station to
Schuylerville and from Schuylerville to Waterford. Thus under baseline conditions, PCB
transport increased downstream consistent with an increasing TSS load. The increase in PCB
loads between TI and Schuylerville are consistent in magnitude with the increase in TSS, i.e.,
TSS and PCB loads both increase about the same amount. However, to Waterford, the increase
in TSS load is significantly greater than the increase in PCB load. The substantially greater
increase in TSS load reflects the solids contributions from the Hoosic River and the Batten Kill.
This relationship between TSS load and PCB load was not evident during Phase 1, as discussed
below.
1-1). 5.2. PCB and TSS Gross Loads under Active Dredging
Table l-D-2 shows PCB homolog and Total PCB load at the three far-field stations for the period
from July 5 through October 27, 2009. The same information is presented graphically in Figure
l-D-3a. As shown in this table and these figures, PCB homolog and Total PCB loads were
s 100
Where:
%Ml
Mff]
Mff2
percent mass loss/gain (unit less)
mass of PCB homolog at the 1st far-field station
mass of PCB homolog at the 2nd far-field station
1-1). 5.1. PCB and TSS Loads under Baseline Conditions
Hudson River PCBs Site Page l-D-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
highest at Thompson Island station (TPCB of 500 kg), and progressively decreased from
Thompson Island to Schuylerville (TPCB of 290 kg), and from Schuylerville to Waterford
(TPCB of 190 kg). However, the percent loss for dichlorobiphenyl, trichlorobiphenyl,
tetrachlorobiphenyl, and pentachlorobiphenyl are greater from Thompson Island to Schuylerville
than from Schuylerville to Waterford. Conversely, the percent loss for monochlorobiphenyl and
hexachlorobiphenyl is greater from Schuylerville to Waterford than from Thompson Island to
Schuylerville. It should be noted that percent mass fraction and mass loss for the
heptachlorobiphenyl, octachlorobiphenyl, nonachlorobiphenyl, and decachlorobiphenyl were not
calculated because the contributions from these homologs are minimal. Table l-D-2 and Figure
l-D-3b show total TSS load at the three stations for the period July 5 to October 27, 2009. As
was observed for baseline conditions, the TSS table and corresponding figure show an increasing
trend from Thompson Island to Schuylerville and from Schuylerville to Waterford, although the
cumulative TSS loads are lower than the baseline TSS loads. However, when the TSS loads
shown in Table l-D-2 are scaled for the time difference between the periods being compared (~4
months vs. 7 months), the results are consistent with the cumulative baseline period average to
within 20 percent or less.
Based on these data, an increasing TSS trend downriver is observed during both baseline and
Phase 1. The increases in solids loads from station to station are generally consistent from
baseline to Phase 1. However, the changes in PCB loads are not, with PCB concentration
increasing downstream under baseline but decreasing under Phase 1.
1-D.6. Discussion of Potential Causative Mechanisms for Observed PCB and TSS Load
Patterns
A summary of the various hypotheses that have been advanced to explain the observed PCB and
TSS transport patterns, along with the associated mechanisms, are discussed below. Baseline
conditions and the active dredging period are discussed separately.
l-D.6.1. PCB and TSS Transport
l-D.6.1.1. PCB and TSS Transport under Baseline Conditions
In the review above, PCB sampling data for years 2004 to 2008 (May to November) revealed an
increasing trend for the PCB homologs between Thompson Island and Waterford. A likely
explanation for the observed increasing trend of the PCB homologs with distance downstream
from Thompson Island appears to be the increase in TSS load, which increases from station to
station downstream, suggesting scour of river bottom sediments between stations. This is
particularly true for the Thompson Island to Schuylerville increase, since there are no major
tributaries in this section. Notably flow is only expected to increase 4 percent based on drainage
area whereas the increases in TSS, Total PCB s and Tri+ PCBs are 27, 14 and 27 percent,
respectively under baseline. This evidence is considered strongly indicative of net sediment
scour in this river section. In the Schuylerville to Waterford segment, there are two major
tributaries known to significantly increase the flow and solids load in this section. However,
these tributaries have been shown to have orders of magnitude lower PCB concentrations in their
sediments and similarly low levels in loads relative to the Upper Hudson (US EPA, 1997). Thus
again, the increase in PCB loads is attributable to resuspension of PCB-bearing sediments from
the river bottom in this river segment.
Hudson River PCBs Site Page l-D-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
1 -D. 6.2. Potential Homolog Transport Mechanisms
Figure l-D-4a shows that the TPCB/Tri+PCB ratio decreases moving downstream from
Thompson Island to Waterford under baseline conditions. From this diagram it can be seen that
the mixture leaving the Thompson Island Pool is roughly 2/3 mono plus di homologs and 1/3
Tri+ PCBs. When the mixture arrives at Waterford, this relationship has changed significantly,
with Tri+ PCBs now comprising nearly 50 percent of the mixture. The drop in the
TPCB/Tri+PCB ratio from Schuylerville to Waterford is greater than the drop in the
TPCB/Tri+PCB ratio from Thompson Island to Schuylerville, correlates to the greater increase
in PCB load that occurs between these stations (see Figure l-D-2a). Examined on an individual
homolog basis, the data on Figure l-D-4b indicate a consistent increase in the PCB load from the
Thompson Island Station to Schuylerville and from Schuylerville to Waterford for all homologs
under baseline, with the exception of monochlorobiphenyls which increased only slightly
between Thompson Island and Schuylerville and then dropped slightly from Schuylerville to
Waterford. Figure l-D-4b also shows that the magnitude of the increase in the PCB load from
Schuylerville to Waterford for most homologs, including dichloro- through hexachlorobiphenyl
is higher than the increase in the PCB load from the Thompson Island Station to Schuylerville.
l-D.6.2.1. PCB and TSS Transport during Active Dredging
Five years of baseline monitoring established a fairly predictable relationship among PCB loads
at the monitoring stations, with both TSS and PCB loads increasing steadily downstream. During
dredging, however, a consistent decline in all PCB homolog loads between Thompson Island and
Waterford was observed, based on the July 5 to October 27, 2009 sampling data despite an
increasing TSS load downstream, consistent in direction and magnitude with the observations of
TSS under baseline. Notably, the observed homolog loss patterns under Phase 1 differed between
the lighter and heavier homologs, as discussed below. The following discussion will examine
possible explanations for these observations.
1 -D. 6.3. Potential Tri+ Homolog Loss Mechanisms
While the Tri+ homologs predominantly exhibited greater mass losses from Thompson Island to
Schuylerville than from Schuylerville to Waterford, the monochlorobiphenyl and
dichlorobiphenyl homolog loss patterns varied between the two river segments. A bar chart of
TPCB/Tri+PCB ratios (Figure l-D-4c) shows an increasing trend in the ratio from Thompson
Island to Waterford, with a greater percentage increase in the ratio between Thompson Island
and Schuylerville (20 percent) than between Schuylerville and Waterford (7 percent). This is
consistent with a greater percentage loss of Tri+ PCBs from Thompson Island to Schuylerville
than from Schuylerville to Waterford during dredging but differs from the increasing downriver
trend in the ratio under baseline, which was attributed to sediment scour and resuspension (see
Section l-D.3.1.1 above). GE hypothesized that the loss of PCBs between Thompson Island and
Lock 5 during dredging is due to setting of particulate matter. However, this hypothesis is not
consistent with the increases in the observed TSS loads between the two stations. As shown in
Table l-D-2and Figure l-D-3b, the TSS load increased by 16 percent from 4,270 kg at
Thompson Island to 4,950 kg at Schuylerville, an increase well within the range observed under
baseline. A more plausible hypothesis is the recycling of particles resulting in dilution of the
Tri+ PCB concentration and a decline in Tri+ load, and in addition, loss of lighter fractions via
gas exchange. It should be noted that plots of TPCB/Tri+PCB ratio versus flow for the three far-
field stations consistently show a decreasing ratio with increasing flow. The referenced plots for
Hudson River PCBs Site Page l-D-6
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
the three stations are provided as Figures l-D-5a, I-D-5b, and l-D-5c. Because sediment scour
and resuspension are more likely under higher flow conditions, particle recycling with less
contaminated sediments in the far-field would maintain the high Tri+ component of TPCB, while
loss of monochlorobiphenyl and dichlorobiphenyl would ensure that the ratio increases
downstream.
While none of these hypotheses can be confirmed, any settling of contaminated particles during
transport from Thompson Island will not significantly change the downstream contaminant
exposure in surface sediment (see Topic 3).
1 -D. 6.4. Potential Lighter Homolog Loss Mechanisms
A 1999 study that was conducted by TAMS Consultants to examine observed PCB loss patterns
titled "Examination of the Gas Exchange Flux for PCBs from the Upper Hudson Schuylerville to
Waterford" (Attachment C of "Data Evaluation and Interpretation Report and Low Resolution
Sediment Coring Report [DEIR LRC] for Hudson River PCBs Superfund Site Peer Review
Responsiveness Summary) reported theoretical air-water gas exchange calculations on the scale
of 40 percent of total PCB inventory for the lighter congeners between Thompson Island and
Waterford. Table l-D-2 shows observed losses for monochlorobiphenyl that that exceed 40
percent of starting inventory, which suggests that other mechanisms besides air-water gas
exchange might be responsible for the lighter (and potentially, heavier) PCB homolog losses.
To explain the additional PCB losses, it can be hypothesized that direct mass losses of PCB Non-
aqueous Phase Liquids (NAPLs) to the atmosphere through air bubbles entrained in turbulent
flow at dams may account for the additional declines in the lighter PCB congeners beyond losses
theoretically attributable to air-water gas exchange. Based on a review of the published
literature, it was found that a void fraction of up to approximately 20 percent is possible near the
plunge pool of cascading flow over a dam where the most turbulence occurs (Chanson et al.,
2002; Hoque, 2002). Assuming 20 percent air entrainment, an average water flow rate of
approximately 5,000 cfs equates to approximately 1,000 cfs air flow comprising PCB NAPL-
coated air bubbles under turbulent flow conditions in the plunge pool downstream of each dam,
potentially resulting in substantial PCB losses to the vapor phase. The following approximate
calculations (based on Aroclor 1242, which is the predominant PCB mixture at the site) provide
a measure of the potential loss rates that might be achievable by this mechanism if full
equilibrium conditions are attained:
Potential PCB Mass Loss Rate
28000 1/sec x 0.001mmHg/760mm Hg x 0.0012 kg/1 = 0.00004421 kg/sec = 3.82 kg/day
Given
Assumed Air flow: 1,000 cfs (28,000 liter/sec
Aroclor 1242 vapor pressure: 0.001 mmHg (source: NIOSH)
1 atm:
Density of air:
760 mmHg
1.2 kg/m3 (0.0012 kg/liter)
Hudson River PCBs Site
EPA Phase 1 Evaluation Report - Addendum
Page l-D-7
The Louis Berger Group, Inc.
April 2010
-------
While this calculation makes several assumptions, it does illustrate the potential for gas exchange
of the PCB oils to be very important under turbulent conditions as the water is transported over
the dams.
1-D.7. Conclusions
In the analyses for the fate and transport of PCBs and TSS within the Thompson Island Pool and
within the Upper Hudson River described above, several hypotheses and mechanisms proposed
by EPA for the changes in the PCB homolog distribution with movement downstream have been
postulated, examined carefully and discussed. These include among others, the formulation of
gas bubbles due to turbulence and those induced by flow over dams into the plunge pool, as well
as gas exchange from the dissolved phase and NAPLs, and the increase in river flows and
suspended solids due to contribution from major tributaries between the far-field stations. These
hypotheses have also been compared with those advanced by GE in their Phase 1 Evaluation
Report (GE, 2010). While some of these hypotheses postulated by EPA can be confirmed by
well established peer-reviewed studies and others cannot, it becomes clear that any settling of
contaminated particles during transport from Thompson Island will not significantly change the
downstream contaminant exposure in surface sediment (see Topic 3). EPA's analysis is
consistent with the entire set of available data.
References
Chanson, H., Aoki, S. and Maruyama, M. 2002. "Unsteady air bubble entrainment and
detrainment at a plunging breaker: dominant time scales and similarity of water level
variations," Coastal Engineering 46(2): 139-157.
EPA, 2010. Phase 1 Evaluation Report. Prepared for EPA Region 2 and USACE Kansas City
District by The Louis Berger Group, Inc. March 2010.
GE, 2010. Phase 1 Evaluation Report. Prepared for General Electric Company, Albany, NY by
Anchor QEA, LLC and Arcadis. March 2010.
Hoque, M.A. 2002. "Air Bubble Entrainment by Breaking Waves and Associated Energy
Dissipation." Ph.D. thesis, Toyohashi University of Technology, Japan.
Richards, R.P. 1998. Estimation of Pollutant Loads in River and Streams: A guidance document
for NPS Programs. Project report prepared under Grant X998397-01-0,
U.S.Environmental Protection Agency, Region VIII, Denver. 108 p.
USEPA, 1997. "Data Interpretation and Evaluation Report - Hudson River PCBs Reassessment
RI/FS, New York." Prepared for the US Environmental Protection Agency Region 2 and
the US Army Corps of Engineers Kansas City District by TAMS Consultants, Inc., the
Cadmus Group and Gradient Corp. February, 1997.
Wagner, L. A. 1982. "Drainage areas of New York streams, by river basins; a stream gazetteer;
Part 1, Data compiled as of October 1980." U.S. Geological Survey, Water-Resources
Investigations, Open-File Report 81-1055
Hudson River PCBs Site Page l-D-8
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 1-D-l
PCB Homolog and TSS Loads at Far-Field Stations under Baseline Conditions
PCB LOAD (kg) and TSS (metric tons) for TID Station - May 15 to November 30
Year
Mono
Di
Tri
Tetra
Penta
Hexa
Hepta
Octa
Nona
Deca
Total PCB
Ratio TPCB/Tri+
TSS
2004
19.7
34.8
15.8
6.7
2.9
0.2
0.0
0.0
0.0
0.0
80.2
3.1
4,700
2005
17.4
34.7
18.7
7.4
2.0
0.1
0.0
0.0
0.0
0.0
80.3
2.8
6,400
2006
30.8
60.6
34.2
13.5
5.1
0.7
0.0
0.0
0.0
0.0
144.9
2.7
16,000
2007
16.6
28.4
12.8
3.9
1.1
0.2
0.0
0.0
0.0
0.0
63.0
3.5
3,700
2008
17.6
30.3
16.4
5.2
1.4
0.1
0.0
0.0
0.0
0.0
71.0
3.1
5,100
Average
20.4
37.8
19.6
7.4
2.5
0.3
0.0
0.0
0.0
0.0
87.9
3.0
7,200
PCB LOAD (kg) for Schuylerville (Lock 5) Station - May 15 to November 30
Year
Mono
Di
Tri
Tetra
Penta
Hexa
Hepta
Octa
Nona
Deca
Total PCB
Ratio TPCB/Tri+
TSS
2004
22.0
44.1
25.7
10.7
4.3
0.4
0.0
0.0
0.0
0.0
107.2
2.6
5,300
2005
17.6
41.5
22.7
9.7
2.8
0.3
0.0
0.0
0.0
0.0
94.7
2.7
8,300
2006
37.5
62.3
40.1
17.7
5.8
1.3
0.0
0.0
0.0
0.0
164.8
2.5
23,000
2007
15.9
29.0
13.4
4.8
1.5
0.2
0.0
0.0
0.0
0.0
64.7
3.3
2,800
2008
13.7
30.8
18.3
6.3
1.7
0.1
0.0
0.0
0.0
0.0
70.9
2.7
6,300
Average
21.3
41.5
24.0
9.8
3.2
0.5
0.0
0.0
0.0
0.0
100.4
2.7
9,100
PCB LOAD (kg) for Waterford Station - May 15 to November 30
Year
Mono
Di
Tri
Tetra
Penta
Hexa
Hepta
Octa
Nona
Deca
Total PCB
Ratio TPCB/Tri+
TSS
2004
15.76
43.82
26.65
16.16
6.00
0.63
0.00
0.00
0.00
0.00
109.0
2.2
31,000
2005
16.56
45.95
35.29
22.24
7.81
1.87
0.04
0.02
0.00
0.00
129.8
1.9
85,000
2006
36.95
84.78
72.25
37.73
14.19
4.68
0.21
0.05
0.00
0.00
250.8
1.9
90,000
2007
12.51
31.77
15.32
6.96
2.20
0.38
0.00
0.00
0.00
0.00
69.1
2.8
83,000
2008
12.05
36.39
21.90
8.88
2.77
0.16
0.00
0.00
0.00
0.00
82.2
2.4
20,000
Average
18.8
48.5
34.3
18.4
6.6
1.5
0.1
0.0
0.0
0.0
128.2
2.1
62,000
Average PCB LOADs (kg) and TSS (metric tins) for Baseline Period May 15 to November 30
Station
Mono
Di
Tri
Tetra
Penta
Hexa
Hepta
Octa
Nona
Deca
Total PCB
Ratio TPCB/Tri+
TSS
TID
20.42
37.76
19.57
7.37
2.48
0.27
0.00
0.00
0.00
0.00
87.90
3.00
7,200
Schuylerville
21.33
41.54
24.03
9.84
3.23
0.46
0.00
0.00
0.00
0.00
100.40
2.70
9,100
Waterford
18.76
48.54
34.28
18.40
6.59
1.54
0.05
0.01
0.00
0.00
128.20
2.10
62,000
Hudson River PCBs Site
EPA Phase 1 Evaluation Report
Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Table l-D-2
PCB and TSS Loads From July 5 to October 27, 2009 for Different Far-Field Stations during Active Dredging
Mass (kg)
Mass Loss (kg)
Percent Mass Loss (%)
Mass Fraction Transport
11 to
bchuylerville
11 to
bchuylerville to
PCB Homologue
TID
Schuylerville
Waterford
Schuylerville
to Waterford
Schuylerville
Waterford
TID
Schuylerville
Waterford
Mono
Monochlorobiphenyl
SO
60
30
20
30
25%
50%
0.16
0.21
0.16
Di
Dichlorobiphenyl
ISO
110
SO
70
30
39%
27%
0.36
0.3S
0.42
Tri
Trichlorobiphenyl
130
70
40
60
30
46%
43%
0.26
0.24
0.21
Tetra
Tetrachlorobiphenyl
70
30
20
40
10
57%
33%
0.14
0.1
0.11
Penta
Pentachlorobiphenyl
30
10
S
20
2
67%
17%
0.06
0.03
0.04
Hexa
Hexachlorobiphenyl
10
7
4
3
3
31%
44%
0.02
0.02
0.02
Hepta
Heptachlorobiphenyl
3
1
0.36
2
1
Octa
Octachlorobiphenyl
1
0.13
0.01
1
0.12
Nona
Nonachlorobiphenyl
0.14
0.02
0
0.11
0.02
Deca
Decachlorobiphenyl
0
0
0
0
0
Total PCB
500
290
190
210
100
Total TSS
4270
4950
3S614
Notes: 1. Numbers are rounded to 1 or 2 significant digits; 2. Percent mass loss and mass fraction transport are not calculated for Mono- through Hexachlorobiphenyl only since the contribution from Hepta- through
Decachlorobiphenyl to the Total PCB are small
Hudson River PCBs Site
EPA Phase 1 Evaluation Report
Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
c
o
1
0.9
0.8
0.7
« 0.6
(/>
to n c
re 0.5
• Far-Field
'PCB Release Mechanism
Near-Field Transect
5 6
Homolog Group
-o
8
o
9
10
Average Mass Fraction for Each Homolog Group
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure 1-D-la
April 2010
-------
Station
# Near-field Release Mechanism
• Moar-fiolrl Trancort QtniHx/
I 1UWI IIVIU I I Wl jr
Thompson Island Dam Automated Station
Values represent mass fraction of Total PCB.
PCB Homolog Patterns Observed in Near-Field and Far-Field
Measurements in TIP
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure 1-D-lb
April 2010
-------
140
120
100
Figure l-D-2a
i TID
Schuylerville
Waterford
O)
T3
ra
o
m
o
CL
80
60
40
20
Mono
70,000
60,000
Di Tri Tetra Penta Hexa Hepta
Homologue Group
Figure l-D-2b
Octa
Nona
Deca
Total
PCB
TID
Schuylerville
Far-field Stations
Waterford
Average Cumulative Homolog, Total PCB Loads and TSS at Far-Field
Stations under Baseline Conditions for May to November (2004 to 2008)
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-D-2
April 2010
-------
200
Figure l-D-3a
Mono
45,000
w 30,000
c
o
+-»
«2 25,000
o
20,000
tn
tn
i TID
Schuylerville
Waterford
Di
Tri
Tetra Penta Hexa Hepta
Homologue Group
Fieure l-D-3b
Octa
Nona
Deca
TID
Schuyerville
Far-field Stations
Waterford
PCB Homolog Loads and TSS Load for July 5 to October 27, 2009
at Far-Field Stations during Dredging
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure l-D-3
April 2010
-------
TID
Schuylerville
Far-field Stations
Waterford
Average Total PCB/Tri+PCB Ratio at Far-Field Stations under
Baseline Conditions for May to November
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure l-D-4a
April 2010
-------
60
50
40
M
00
u
Q.
20
10
TID
Schuylerville
Far Field Station
Waterford
Average of PCB Loads Measured at Far-field Stations during
Baseline Period (2004 - 2008)
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure l-D-4b
April 2010
-------
3.00
2.50
2.00
1.50
1.00
0.50
0.00
TID
Schuylerville
Waterford
Ratio TPCB/Tri+
TPCB/Tri+ Ratio for July 5 to October 27, 2009
at Far-Field Stations during Dredging
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure l-D-4c
April 2010
-------
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
~ ~
~
~
A
A A
y =-0.0002x + 3.3024
R2 = 0.4008
1000 2000 3000 4000 5000 6000
Fort Edward Flow (cfs)
7000
8000
9000
10C
TPCB/Tri+PCB Ratio vs. Fort Edward Flow at Thompson Island
During Dredging
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
-------
c
s
* 3
pq
u
CL.
+
H
pa
y 2
o
H
y =-0.0002x + 3.8488
R2 = 0.4338
2000
4000 6000
Fort Edward Flow (cfs)
8000
10000
12000
TPCB/Tri+PCB Ratio vs. Fort Edward Flow at Lock 5
During Dredging
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure l-D-5b
April 2010
-------
Waterford Station
s
s
«
Pi
pp
u
a.
+
H
pq
u
a.
o
H
¦
¦
¦
¦
¦
¦
¦
¦ ¦
V
¦
¦
¦
lb-
_ ¦
¦
I"
¦
¦¦ ¦¦ ¦
¦¦ ¦¦ _
¦¦ i1 Mi
m ¦ *
-y- . ¦ ¦¦
¦ « ¦
¦_
VST
¦ i*Tkrf
'¦ ¦ ¦
¦ ¦
¦
.
¦
¦ -¦¦¦ I"
¦ ¦ ¦ Hi, 1
-
%
¦¦ ¦
¦
¦ ¦
¦ 1
.
^ ¦
¦
¦
¦
¦
¦
y =-0.000 lx + 3.7096
R2 = 0.2316
5000
10000 15000
Fort Edward Flow (cfs)
20000
25000
30000
TPCB/Tri+PCB Ratio vs. Fort Edward Flow at Waterford
During Dredging
EPA Phase 1 Evaluation Report - Addendum- Hudson River PCBs Site
Figure l-D-5c
April 2010
-------
c4
at
CO
u
a.
+
CO
u
a.
o
H
A Thompson Island
Lock 5
¦ Waterford
X Near-field Transect
X Near-field Release Mechanism
(18.35®. 10/20 for
Near-field Transect)
X
X*
11111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111
in in in c— c— c— c— oooooo c^c^c^^-<^-
-------
TOPIC 2
CAUSES OF RESUSPENSION
DURING PHASE 1 DREDGING
-------
Topic-2 -Causes of Resuspension during Phase 1 Dredging
Abstract
Water column PCB concentrations at Thompson Island Dam (TID) were compared with daily
measurements of dredging and backfilling processes to develop multiple regression models
(Kern Statistical Services, Inc. 2010). The analysis concludes that the mechanisms associated
with increased water column PCB concentrations are varied and likely many and should not be
simplified to a simple proportionality to mass or volume removed. Rather, metrics quantifying
Phase 1 dredging are the most promising guide to how dredging and other activities could be
modified to reduce resuspension of PCBs in Phase 2. Using a combination of Factor Analysis
and Multiple Regression of the daily process and water column data, EPA has determined that
process metrics are positively associated with and influence release and resuspension of PCB
contamination. These metrics include:
1. Sediment removal {i.e., bucket counts, volume removed, mass removed),
2. Flow rate,
3. Vessel traffic (primarily distance traveled by scows pushed by tugs),
4. The number of certification units (CUs) being backfilled in any given day,
5. The area and concentration of disturbed sediments in CUs open to the water column, and
6. Bucket fill-rate and other surrogates to PCB contaminated sediment and water spillage.
Given EPA's understanding that the primary drivers of resuspension are associated with the
collective group of 6 process variables identified above, it is reasonable to expect that if these six
process variables are adjusted to levels observed during periods of resuspension less than 200ng/l
and productivity on the order of 3,000 to 3,500 cubic yards per day (August 9 to September 7,
2009 and September 11 to October 5, 2009), future resuspension results would be improved over
the Phase 1 performance. This analysis also indicates that Phase 2 dredging can be conducted in
a way that results in reasonable resuspension rates and also meets productivity goals.
2.2.Introduction
In their Phase 1 Evaluation Report, GE has indicated that:
1. Phase 1 demonstrated that the Resuspension Standard cannot be met (ES-5), and that EPA's
proposed increase to the load standard would compromise the remedy (ES-7).
2. The federal drinking water standard of 500 ng/L PCBs will be exceeded frequently in Phase
2 because of higher rates of dredging and fewer resuspension control opportunities than in
Phase 1.
3. Re-deposition in non-dredge areas will compromise remedy benefits.
4. There are no practical means to reduce resuspension to avoid exceeding the Resuspension
Standard.
Further, GE has adopted the view that the apparent relationship between dredged volumes of
sediment and resuspension is inevitable. Based on recent analysis (Kern Statistical Services, Inc.
2010) included as Attachment 2 to the Addendum, it can be concluded that this view is over-
Hudson River PCBs Site Page 2-1
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
simplified and that the mechanisms associated with increased water column PCB concentrations
are varied and, likely, many and should not be reduced to a mere proportionality to mass or
volume removed, as suggested by GE. EPA agrees that mass or volume removed is a surrogate
for the net effect of all of the processes involved in dredging, and therefore correlates well with
water column PCB concentrations. However, this does not preclude that individual operational
variables can be managed to reduce resuspension of PCBs. The manner in which dredging and
other related activities were conducted during Phase 1 is very important and, therefore, provides
a good source of information on how dredging and other activities could be modified to reduce
resuspension of PCBs in Phase 2.
EPA's analysis of factors associated with water column PCB concentrations has found that while
water column PCB concentrations are indeed positively associated with mass of PCBs removed,
a more careful analysis indicates that this relationship is due to a combination of several
operational factors, some of which are readily manageable in ways that would logically be
expected to reduce PCB releases associated with dredging operations. EPA also observes that
there were periods of time in Phase 1 when resuspension rates were low (less than 200 ng/1, and
productivity was adequate to meet proposed Phase 2 productivity criteria.
Based on a multivariate analysis of the daily process and water column data, EPA finds that
water column PCB concentrations are positively associated with several factors, all of which
would be expected to influence release and resuspension of PCB contamination, including:
1. Sediment removal {i.e., bucket counts, volume removed, mass removed),
2. Flow rate,
3. Vessel traffic (primarily distance traveled by scows),
4. The number of CUs being backfilled in any given day,
5. The area and concentration of freshly disturbed sediments in CUs open to the water column
each day, and
6. Bucket fill-rate, cycle time and other surrogates to sediment spillage.
Thirteen of the 28 process variables (shown in Tables 2-1 through 2-3) considered by EPA
demonstrated statistically significant positive associations with water column PCB
concentrations with squared Spearman Rank correlation coefficients ranging from approximately
0.2 to 0.4. These levels of association are individually weak, indicating that no single process
can be identified as "the source" of resuspension; but rather a complex set of interactions among
processes appears most likely to be "causativeThe well known adage is that "association does
not imply causation" but, conversely, lack of apparent association also does not eliminate
causation. It is expected that multiple variable models may be necessary to adequately explain
variation in water column concentrations. It is further expected that controlling PCB
resuspension may be accomplished through a combination of additional or improved
management practices / strategies applied to several stages in the sediment removal and
backfilling process.
Hudson River PCBs Site Page 2-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
2.3.0bjective
The primary objective of EPA's analysis is to develop an empirical data-driven model describing
water column concentrations of total PCB as a function of physical and operational variables
associated with remedial activities in the Phase 1 project.
2.4.Method
Data specific to individual CUs from the Phase 1 project provides the basis to develop an
empirical model of water column concentration as a function of measured data collected on a
daily basis specific to the operations in the Thompson Island Pool in 2009. A combination of
factor analysis (Seber, 1977) and multiple regression (Neter et al., 1996) is used to statistically
identify groups of parameters most strongly associated with water column PCB concentrations.
This empirical approach provides the opportunity to test hypotheses and assumptions that
otherwise would remain untested in more typical situations where data are less rich—providing
the means to evaluate the influence of various remediation processes on water column
concentration.
The data include metrics quantifying potential sources of PCBs associated with:
1. Flow and temperature conditions,
2. Debris removal,
3. Volume and mass removed,
4. Vessel traffic,
5. Efficiency of removal operations,
6. Resuspension of exposed PCB deposits in open CUs, and
7. Sediment disturbance associated with backfilling.
In EPA's analysis, 28 variables were tested statistically for potential to predict water column
PCB concentration at far field stations downstream of dredging operations. Data were
summarized on a daily basis, so for most variables there were approximately 166 days (May 15
through October 27, 2009) on which PCB concentration could be compared with dredging
process variables. Some variables such as bucket fill rates were only available on the 127 days
when active dredging occurred within the 166 day time period. Therefore the analyses were
repeated for variables measured on all 166 days as well as for the subset of variables measured
on just 127 days, primarily after June. Data measured on the smaller subset of days are more
closely associated with how dredging was conducted and are therefore the more likely source of
information on how dredging and other activities could be modified to reduce resuspension of
PCBs to the water column.
2.5.Modeling Overview
This factor analysis and regression approach is used to develop a model of the form
C,
water
Hudson River PCBs Site
EPA Phase 1 Evaluation Report - Addendum
Page 2-3
The Louis Berger Group, Inc.
April 2010
-------
Where the constants Kj, K2, K3,... Kn are loosely interpreted as "nef sediment-to-water
partitioning coefficients for each source, and where it is understood that source terms are based
on surrogates for sediment removal processes.
The Phase 1 project is unique in the richness of data available to not only estimate these
coefficients, but also to identify those combinations of measured processes that are most
important for predicting water column concentrations. A combination of factor analysis and
multiple regression analysis is used to identify metrics contributing significantly to prediction of
water column PCB concentrations. These are further described in Attachment 2.
2.6.Results - Bivariate Correlations
EPA identified thirteen variables representing 5 groups of processes that were associated with
water column PCB concentrations at the Thompson Island Dam monitoring station (TID). Five
variable groups (factors) were identified that collectively explained 55 and 60 percent of the
variation in water column PCB concentration in the 166-day and 127-day models, respectively.
These factors represented volume and mass removed and efficiency, area of recently disturbed
sediments in open certification units, vessel traffic and backfilling. Following is a summary of
the results of the analysis.
Squared Spearman Rank correlation coefficients for water column PCB concentration with each
of the process variables are reported in Table 2-1. The analysis was repeated with process
variables paired with one-day and two-day lagged PCB concentrations to evaluate the potential
effects of travel time on the strength of correlation. These squared correlation coefficients
represent the proportion of variation explained by the relationship between water column PCB
concentration and each variable, analogous to an R2 from a regression. The Spearman
correlation coefficient is preferred because the assumption of linearity inherent in the Pearson's
coefficient is relaxed. Results summarized in Table 2-1 show that:
1. Correlations between water column PCB concentrations and process variables are generally
weak, ranging from 2 percent for debris removal to 42 percent for volume and mass removal,
indicating that no single variable could be expected to adequately explain the fluctuations in
water column PCBs observed during the Phase 1 project.
2. Correlations for water column concentrations that were lagged by one day were less than
those for concurrent measurements and two-day lagged measurements produced still lower
correlations.
a. In contrast GE asserted that weekly averages were needed to counter the effects of travel
time in their analysis of the water column PCB data.
b. EPA views this as counterproductive given the lack of correlation between lagged water
data and process variables.
c. Weekly averaging would artificially reduce the power to detect subtle multiple variable
relationships, thereby suppressing our ability to identify potentially important
relationships between water column PCB concentration and operational variables.
3. Statistically significant positive associations were identified for most processes expected to
disturb sediments
Hudson River PCBs Site Page 2-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
a. Volume and mass removed (R2=0.22 to 0.42)
b. Dredging efficiency measures such as bucket fill rate and depth of cut (R2=0.08 to 0.22)
c. Sources due to area of open CUs (R2=0.15 to 0.19)
d. Debris removal (R2=0.02)
e. Boat traffic (R2=0.11 to 0.26)
f. Backfilling operations (Number of CUs being backfilled) (R2=0.09)
4. Weak statistical relationships may be indicative of surrogate relationships that are markers
for important, but crudely quantified, sources of PCB resuspension.
5. Water column concentrations were negatively associated with flow at the Fort Edward
Station, but the relationship was not statistically significant.
2.7.Regression on Factor Scores - Factor Analysis
Because water column PCB concentrations were weakly associated with several individual
operational variables, EPA developed a multivariable model that would adequately explain water
column PCB concentrations. In order to understand interrelationships between process variables
a factor analysis was conducted to identify a set of independent factors that would be
meaningful, as well as providing inputs for a regression model predictive of water column PCB
concentrations.
The factor analysis was conducted with only the predictor variables for the subset measured on
all 166 days, as well as the subset measured on just 127 of the 166 days. Resulting factor scores
were used as predictors in a regression analysis to identify important factors for prediction of
water column PCB concentrations.
2.7.1. Factor Analysis (127-day model)
Table 2-2 shows the factor loadings for each of the 28 process variables under consideration.
The factor loadings are unitless and range from plus one to minus one and are considered
meaningful when they exceed approximately 0.4 in magnitude. Cells in Tables 2-2 and 2-3 are
shaded green to draw attention to factor loadings that exceed this nominal level.
There were 5 factors associated with total PCB concentration in the water column describing
from 2 to 37 percent of the total 60 percent variance in water column PCB concentration
explained by the regression model.
Factor-1 includes loadings on bucket counts, mass removed, volume removed, residual Total
PCB Concentration in Open CUs and the product of mass and removal efficiency (ME). Factor-1
summarizes potential PCB sources from variables that are directly related to sediment removal,
as well as efficiency of the removal process.
Factor-6 is most heavily weighted on the amount of backfilling being conducted and the product
of flow and backfill (a surrogate for load from backfilling). Factor-6 also has substantial negative
loadings on bucket counts and temperature. This may reflect that bucket counts coincidentally
vary inversely with temperature and backfilling operations {i.e., it is coincidental that dredging
started slowly when temperatures were colder, increased through the summer and then declined
in the fall when temperatures also declined and backfilling commenced).
Hudson River PCBs Site Page 2-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Factor-7 is most heavily weighted on concentration-weighted surface area of open CUs and the
product of flow and concentration-weighted surface area of open CUs. Factor-7 has a clear
signal exclusively related to the amount of open CUs at any point in time that is independent of
volume and mass removal.
Factor-8 is most heavily weighted on flow, and the product of flow and total vessel traffic.
Factor-8 is also independent of variables in Factor-1 describing removal metrics, indicating that
there may be an independent PCB source to the water column associated with vessel traffic.
This variable is a crude measure of potential sources due to vessel traffic, because it does not
account for either water depth or concentration of areas over which traffic occurs. It is expected
that refinement of this variable will substantively improve its relative strength as a predictor of
water column concentrations.
Factor-9 is most heavily weighted on boat distance, which is a single metric that only accounts
for distance traveled by vessels.
As a general observation, these results show that resuspension of PCBs to the water column is
associated with a combination of removal activities, backfilling activities, vessel traffic and the
surface area and duration that disturbed residuals are exposed in open CUs. This indicates that
additional or improved management practices could be applied to one or several of these
processes to reduce concentrations of PCBs in the water column.
2.7.2. Factor Analysis (166-Day Model)
The factor loadings for the 166-day model are summarized in Table 2-3. Results were similar to
those for the 127-day model because the majority of data were common to both models. The
model fit was slightly weaker with an adjusted R2=0.55 as compared with 0.60 for the 127-day
model. The five factors were qualitatively similar to those identified in the 127-day model
representing variables associated with sediment removal (semi-partial R2=0.28), backfilling
(semi-partial R2=0.06), concentration-weighted surface area of open CUs (semi-partial R2=0.04),
flow times vessel distance (semi-partial R2=0.05), and mass removed (semi-partial R2=0.13).
Because performance data related to bucket filling rates were not included in the 166-day model,
the separation of variables among factors was less obvious and general surrogates for overall
activity such as mass and volume removal and boat traffic tended to group together in the first
factor. This shows that further refinements in the understanding of processes controlling fluxes
of PCBs to the water column should focus on variables that characterize how dredging and other
supporting operations are conducted as opposed to just on how much dredging is done.
2.7.3. Regression on Mechanistic Variables
Model fit was improved through inclusion of mechanistic variables that were summarized
specific to individual CUs, increasing the adjusted R2 for the 127-day model from 0.60 to 0.68.
The primary improvements in model fit were due to incorporation of the dredge efficiency term
in place of the more general mass removed term proposed by GE, and also importantly by
differentiating vessel traffic by water depth and sediment bed concentration, and finally by
incorporation of a source term specific to activities that occurred during dredging performed
inside the sheet pile wall at CU-18, particularly in late July and early August.
Hudson River PCBs Site Page 2-6
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
2.8.Primary Mechanisms Influencing Water Column PCB Concentrations
A multiple regression model for water column PCB concentration was developed using
individual measured variables rather than factors as described above. The model included 6
variables and the overall proportion of variation explained by the model was 0.68 and the
proportion explained by each variable was: mass removal efficiency (partial R2=0.25); boat
traffic specific to backfilling operations (partial R2=0.11); boat traffic specific to mass removal at
CU-17 and CU-18 (partial R2=0.01); flow at the Fort Edward station (partial R2=0.08); the
number of scows on queue for unloading (partial R2=0.01) and mass removed from inside the
sheet pile wall at CU-18 (partial R2=0.03). When predictors are inter-correlated, the sum of the
variance explained by each variable independently is often less than the total variance explained
{i.e., adjusted R2). This component of the total variance is the variance explained jointly by the
collection of variables in the model. In this case 19 percent of the variance was explained jointly,
indicating that approximately a quarter of the variance explained by the model cannot be
uniquely tied to any particular factor. These results are summarized graphically in Figure 2-1.
Figure 2-2 shows the fitted model plotted against time, along with the observed water column
concentrations and upper 95 percent confidence limits for predicted PCB concentrations. This
model based on mechanisms provides better agreement between observed and predicted values,
particularly during early August and early September when there were high excursions in the
TID PCB concentrations.
The mass removal efficiency variable was the primary predictor of water column concentrations
explaining 24 percent of the total variance in water column PCB concentrations. This variable is
comprised of the product of mass removed and apparent bucket fill rate. This variable identifies
situations where mass removal is conducted when daily bucket fill rates are calculated as greater
than 100 percent (thought to represent situations where multiple bucket closures are counted as a
single bite due to debris or less-well controlled operations). EPA evaluated whether this
correlation between dredging efficiencies and water column concentration could be due
primarily to correlation with mass removed. This hypothesis was tested by comparing model fit
for a model including mass removed alone with a model including mass removal efficiency. The
model including the other four variables identified above, plus mass removal efficiency, had an
adjusted R2=0.68, while the model including the same four variables and just mass removed {i.e.,
without multiplying by efficiency) had an adjusted R2=0.60. This indicates that mass removal
efficiency was a better predictor of water column concentration than mass removed alone. This
relationship indicates that, in addition to simple proportionality of mass removed, unclosed
dredge buckets (and potentially other inefficiencies) also contributed to resuspension of PCBs to
the water column.
Disturbance due to tug traffic in shallow areas between CUs 17 and 18 and the Lock 7 entrance
was found to contribute significantly to water column concentrations (semi-partial R2=0.11). A
plot of water column PCB concentration vs. tug traffic weighted by depth and concentration is
given in Figure 2-3. This term might also be thought to be a surrogate for mass removed from
CUs 17 and 18; however, it was found that mass removal in CUs 17 and 18 (outside sheet pile)
was poorly correlated with water column PCB concentrations. This suggests that the tug traffic
associated with removal of 2600 kg of PCBs from CUs 17 and 18 (outside the sheet pile
Hudson River PCBs Site Page 2-7
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
containment) caused erosion and resuspension of PCB-contaminated sediments in areas with
water depths less than 11 feet.
Dredging declined through October, ending on October 27, yet water column concentrations
remained elevated at TID. Disturbance due to shallow-water tug traffic primarily servicing
backfilling operations was associated with water column concentration (partial R2=0.11).
Because the act of backfilling itself is unlikely to cause resuspension of PCBs on the order of
dredging, this relationship represents strong evidence that tug traffic in shallow areas caused
disturbance and resuspension of PCBs to the water column. Water column PCB concentrations
were on the order of 300 ng/L during October when nearly all sediment disturbance was caused
by backfilling and capping operations. These water column PCB concentrations are similar to
those observed during June and July 2009 when most disturbance was associated with dredging-
related activities, including tug traffic. Backfilling itself should be a relatively innocuous activity
because clean fill was placed on sediments that generally had much lower PCB concentrations
than the material removed. However, capping was performed over inventory due to schedule
constraints, in some cases with elevated concentrations, but disturbance of those areas should be
less than occurs with dredging. This indicates that tug traffic was the likely cause of disturbance
leading to resuspension of PCBs during October 2009, as opposed to merely acting as a surrogate
for mass removal.
Tug traffic was associated with water column PCB concentration during both backfilling and
dredging operations. If tug traffic were merely a surrogate for mass removal as opposed to a
causative agent, one would not expect to see the tug traffic signal in the water column record
during backfilling. This indicates that resuspension due to tug traffic is likely to be a true
causative factor of resuspension as opposed to a mere surrogate. Based on these findings, GE
should consider alternative processes for distribution of cap and backfill material that minimize
tug traffic.
Improvement in model fit during the late July to early August period was due to inclusion of the
variable quantifying mass removed in CU-18 from within the sheet pile wall. As these activities
were conducted within the confines of the sheet pile, this is counter-intuitive, but it is thought
that this spike in water column concentrations, particularly during late July and early August,
may be due to the release of extremely highly concentrated PCB contaminated water associated
with accidental losses and bucket decanting as sediment (and water) was lifted over the sheet pile
and into scows. For example, in a letter to EPA dated August 14th, 2009, GE reported water
column PCB concentrations within the sheet pile of 12,800 ng/1, 116,000 ng/1 and 19,800 ng/1, on
July 31, August 5 and August 8, respectively. This was during the period of time when dredging
activities were conducted within the sheet pile and when water column PCB concentrations
spiked at TID.
Dredging in CU-18 was initiated in late July 2009 and at this point in time water column
concentrations increased dramatically at TID to some of the highest levels observed during the
Phase 1 removal. Although there was much less PCB mass produced in CU-18 (inside the sheet
pile) (430kg) in comparison to CUs 17 and 18 (outside the sheet pile) (2600 kg), and in spite of
the fact that the sheet pile wall was in place to contain resuspended PCBs, mass removal in this
CU was strongly associated with water column PCB concentrations. PCB concentrations in
water inside the sheet wall were on the order of 20,000 ng/L and exceeded 100,000ng/l on at
Hudson River PCBs Site Page 2-8
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
least one occasion. In late August when some sheet piling was removed, and dredge equipment
was moved in and out of CU-18, PCB concentrations exiting the sheet pile area were on the
order of 2,000 ng/L. The response in PCB concentrations at TID was nearly immediate,
particularly in late July and Early August and much greater in magnitude than would have been
expected, given the small mass of PCB s removed from inside the sheet pile wall, and the fact
that flow rates were apparently restricted by the sheet pile. Water column PCB concentrations at
TID during these episodes were higher than at any other time during the Phase 1 removal,
indicating that processes other than mere proportionality to mass removed are important to
explain PCB resuspension to the water column.
Associations between dredging activities at CU-18 and water column PCBs were statistically
significant, in spite of the fact that activities inside the sheet pile at CU-18 were of relatively
short duration and represented a very small fraction of the mass removed in Phase 1. While this
operation was relatively small, its influence was disproportionately large; the related activities
represented 3 percent of the total variance explained in water column concentrations over the
entire dredging and backfilling season. Conversely, nearly 6 times as much PCB mass was
produced from CUs 17 and 18 (outside the sheet pile wall), and yet the response in water column
PCB concentrations was minimal after adjusting for the other factors included in the model. This
shows that combinations of other factors better describe water colum PCB concentrations than a
simple proportionality to mass removal rates. Otherwise, general dredging activities at CU-17
and areas outside the sheet pile at CU-18 should have been more strongly correlated with water
column concentrations than the much smaller mass removal from inside the CU-18 sheet pile.
Flow at the Fort Edward Gauge was also included in regression models and was found to be an
important predictor of PCB concentration at TID. Because water column PCB concentrations
were found to be correlated with flow it is important that any analyses attempting to explain
variation in water column PCB concentrations during dredging should explicitly account for
variations in flow. Regression models between loads and flow are not valid for inference to
causative mechanisms, because load is a function of flow. It is not valid to "regress" a function
of flow against flow, because the 1 to 1 correlation of flow with itself is sure to cause a false but
apparent correlation between the functions of interest.
2.8.1. Model Predictions
The fitted model results are plotted on Figure 2-4 showing that the modeled concentrations
generally track day to day fluctuations in concentration in most months, including patterns
observed in October that were not well described by GE's simpler model. This shows that GE's
assertion that concentrations are driven exclusively by the amount of dredging is not justified.
Also included in the plot are upper 95 percent prediction limits which are an added benefit of the
regression approach to model development. It can be seen that the prediction intervals capture
the majority of observations.
2.8.2. Process Settings for Phase 2
This analysis identifies 6 parameters that collectively explain a large proportion of the variation
in water column PCBs at TID. There were periods of time when dredging and backfilling
produced PCB concentrations at TID below 200 ng/1 (August 9 to September 7, 2009 and
September 11 to October 5, 2009) and yet productivity was on the order of 2,500 to 3,500 cubic
Hudson River PCBs Site Page 2-9
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
yards per day (Table 2-4). Within the August period of time active dredging occurred on 21
days, and productivity averaged 2,557 cubic yards per day, exceeding 3,378 cubic yards per day
(EPAs proposed average daily productivity rate) on the three most productive days. There were
also periods of time when PCB concentrations exceeded 300 ng/1 on average (July 10 to August
9, 2009 and October 5 to October 27, 2009). Figure 2-5 compares the distribution of each of the
6 key process variables during these periods of time when resuspension was high with those
periods when resuspension was low. The clear shift in distributions illustrates the magnitude of
difference in how process variables were "set" during Phase 1. This shows that Phase 1 dredging
was at times conducted in such a way that Productivity and Resuspension Standards were met
simultaneously. For Phase 2 dredging, process variables should be targeted to meet the ranges
identified in Figure 2-5 to achieve high productivity with the expectation of low frequency of
resuspension exceedances. As noted in development of the performance standards, flow is an
important predictor of concentrations and extreme flow events may precipitate higher water
column levels, necessitating adaptive management actions.
2.9.Conclusions
The multiple variable factor analysis and corresponding regression is a first step in understanding
the processes contributing PCBs to the water column. Several processes may be contributing to
the PCB loads to TID far field station and there is potential to improve dredging processes while
maintaining a high likelihood that EPA's proposed Resuspension Standard for Phase 2can be
met. The most likely factors contributing PCBs to the water column are not unexpected—mass
and volume removal, vessel traffic, exposure of freshly disturbed residual sediments to active
flows, processes associated with backfilling, and the extent to which dredging may have
encountered debris or have been less-well controlled, resulting in buckets that weren't fully
closed and increasing the potential for PCB losses.
This analysis shows that a combination of processes are likely contributing measureable
concentrations of PCBs to the water column, which presents an opportunity to fine tune dredging
operations in Phase 2.
The factor-based model reported here supports the hypothesis that sources of PCBs to the water
column are many and varied and that there are likely to be opportunities to minimize PCB
resuspension during the upcoming Phase 2 dredging project. GE's statements in their Phase 1
evaluation report are not a reasonable response to the extensive information that is available to
further refine and optimize the dredging operation. Regression of water column PCB
concentration on mechanistic variables extends the multivariate factor analysis toward the goal
of a predictive model than can be used to adjust dredging process parameters with an expectation
of a commensurate response in water column PCB concentrations.
This analysis indicates that primary drivers of resuspension are associated with the collective
group of 6 identified process variables. It is reasonable to expect that if these six process
variables are adjusted to levels generally observed during periods of acceptable resuspension and
productivity, future resuspension results would be improved relative to 2009 performance.
EPA recognizes that there are many process variables that may influence dredging performance
that have either not been measured or, of those measured, did not result in discernable
associations with TID water column concentrations. This analysis shows that resuspension and
Hudson River PCBs Site Page 2-10
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
productivity standards were met simultaneously at times during Phase 1—therefore, it is possible
to do so. It is also fully expected that a relatively wide range of combinations of process best
practices could be used to achieve similarly reasonable resuspension rates at levels of production
necessary to meet the Phase 2 Productivity Standard.
There may also be gross process changes in dredging and backfilling technologies that were not
evaluated in Phase 1 that may achieve similar or larger reductions in resuspension than would be
achieved through incremental changes in existing technologies. For example, this analysis
identified tug traffic associated with both sediment removal and backfilling as being associated
with water column concentrations at TID (Figure 2-3). Backfilling and capping material was
transported to CUs using tugs and scows, so the majority of this tug traffic could be completely
eliminated if backfill and cap material were pumped as slurry and applied hydraulically, rather
than mechanically.
This analysis also supports the conclusion that Phase 2 dredging can be conducted with
reasonable resuspension rates while meeting productivity goals.
References
General Electric, 2009. Letter to David King, USEPA dated August 14th, 2009. From Timothy
Kruppenbacher.
Kern Statistical Services, Inc. 2010. Conditions associated with water column PCB
concentrations: Thompson Island Dam 2009 multiple analysis of water column PCB and
operational data.
Neter, J., Kutner, M.H., Nachtsheim, C.J., and W. Wasserman. 1996. Applied Linear Statistical
Models, 4th ed., Irwin. Chicago.
Seber, G.A.F. 1977. Multivariate Observations. Wiley Series in Probability and Mathematical
Statistics. John Wiley and Sons. New York, NY.
Hudson River PCBs Site Page 2-11 The Louis Berger Group, Inc.
EPA Phase 1 Evaluation Report - Addendum April 2010
-------
Table 2-1.
Squared Spearman rank correlation coefficients between water column total PCB concentrations
lagged by 0, 1 and 2 days. Correlations are more often the strongest when based on concurrent
measurements of water column PCBs and sediment disturbance and productivity factors.
Variable
PCB_ng/L
PCBng/LLagl
PC B_ng/L_Lag2
BargeDist
0.001
0.000
0.003
Barge V_D
0.005
0.004
0.001
BargeVel
0.003
0.002
0.001
BCntTotal
0.217
0.174
0.094
BoatDist
0.338
0.315
0.189
BoatVD
0.315
0.313
0.188
BoatVel
0.275
0.254
0.145
Debris
0.135
0.165
0.186
DrdgDist
0.033
0.016
0.002
DrdgVD
0.004
0.000
0.001
DrdgVel
0.001
0.002
0.010
Fill Rate
0.095
0.063
0.021
FlowFE
0.014
0.035
0.044
Load Bfill
0.091
0.094
0.094
Load_CU_Area
0.191
0.190
0.193
Load_MassRem2
0.379
0.316
0.191
LoadBoats
0.303
0.253
0.136
MassRemTotaB
0.417
0.323
0.185
ME
0.384
0.346
0.225
SbDist
0.108
0.066
0.026
SbVD
0.118
0.076
0.038
SbVel
0.022
0.007
0.000
ScowDist
0.265
0.220
0.140
ScowV_D
0.257
0.225
0.144
ScowVel
0.178
0.125
0.072
TempC
0.007
0.011
0.013
TotalBfill
0.090
0.094
0.093
tPCB CU AREA
0.146
0.147
0.148
VolRemTotal
0.346
0.285
0.161
Notes:
1) Gray cells indicate when water current water column concentrations correlate more strongly than lag-1
measurements or when lag-1 measurements correlated more strongly than lag-2 measurements.
2) Bold numbers indicate that correlations are significantly different from zero at the 5% level of
significance.
3) Number of days represents the number of paired observations for values measured concurrently.
Sample sizes associated with one and two day lags are reduced by one or two respectively.
Hudson River PCBs Site
EPA Phase 1 Evaluation Report - Addendum Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Table 2-2.
Factor loadings for each variable for those factors found to be associated with water column Total PCB concentration in Thompson Island
Pool from May to November in 2009, Hudson River. Loadings range from minus 1 to plus 1 and values greater in magnitude than 0.4
(green shaded and bold) are thought to be meaningful. Based on 127 day model.
Variable
Factorl
Factor6
Factor7
Facto r8
Factor9
BCntTotal
0.66
-0.40
0.19
-0.07
0.13
BargeDist
-0.09
0.22
-0.10
-0.04
0.08
BargeVD
0.02
0.06
-0.04
-0.10
0.09
BargeVel
-0.10
0.22
-0.11
0.00
0.04
DrdgDist
0.04
0.08
0.02
0.01
0.11
DrdgVD
0.03
0.00
-0.05
-0.03
0.04
DrdgVel
-0.11
0.03
-0.12
0.01
-0.08
LoadBfill
-0.16
0.94
-0.03
0.16
0.06
FlowFE
-0.35
0.12
0.00
0.86
-0.19
TempC
0.18
-0.72
0.16
-0.11
0.09
LoadCUArea
0.27
-0.09
0.87
0.24
0.05
MassRemTotal
0.92
-0.11
0.10
-0.11
0.06
SbDist
0.10
0.05
0.02
0.02
0.12
SbVD
0.09
0.13
0.02
-0.01
0.11
SbVel
-0.02
-0.03
-0.11
0.02
-0.13
ScowDist
0.34
-0.03
0.14
-0.03
0.20
ScowVD
0.26
0.04
0.19
-0.02
0.10
ScowVel
0.28
-0.13
-0.15
0.04
-0.05
TotalBfill
-0.14
0.93
-0.05
0.03
0.09
VolRemTotal
0.82
-0.10
0.14
-0.04
0.08
tPCBCUARE
0.44
-0.17
0.78
-0.18
0.13
TotalEfficiency
0.32
-0.10
0.16
0.09
0.03
ME
0.92
-0.11
0.10
-0.06
0.04
BoatDist
0.36
0.15
0.11
0.04
0.70
BoatVel
0.19
0.03
0.06
0.06
0.14
BoatVD
0.36
0.09
0.14
0.03
0.68
LoadBoats
0.01
0.32
0.15
0.77
0.41
Semi-Partial R2
37%
10%
2%
10%
2%
Factor Label
Volume/Mass
Bucket Fill
Backfill and Flow
Weighted Backfill
PCB/Flow Weighted
CU Area
Flow Weighted Vessel
Dist.
Vessel
Distance/Velocity
Hudson River PCBs Site
EPA Phase 1 Evaluation Report - Addendum Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Table 2-3.
Factor loadings for each variable for those factors found to be associated with water column Total PCB concentration in
Thompson Island Pool from May to November in 2009, Hudson River. Loadings range from minus 1 to plus 1 and values
greater in magnitude than 0.4 (green shaded and bold) are thought to be meaningful. Based on 166 day model.
Variable
Factorl
Factor4
Factor6
Factor 7
Factorl2
BCntTotal
0.84
-0.23
0.08
-0.04
0.05
BargeDist
0.08
0.22
-0.12
-0.02
-0.01
Barge V_D
0.12
0.06
-0.04
-0.09
0.04
BargeVel
0.05
0.21
-0.12
0.02
-0.03
DrdgDist
0.18
0.09
0.00
0.01
-0.02
DrdgVD
-0.02
0.00
-0.02
-0.04
0.08
DrdgVel
-0.16
0.00
-0.14
0.03
-0.07
Load Bfill
0.06
0.97
-0.03
0.11
-0.03
FlowFE
-0.45
0.11
-0.03
0.86
0.02
TempC
0.32
-0.65
0.18
-0.12
-0.04
Load_CU_Area
0.32
-0.07
0.88
0.17
0.01
MassRemTotal3
0.84
-0.09
0.08
-0.12
0.44
SbDist
0.42
0.09
0.00
0.03
-0.01
SbVD
0.36
0.16
0.03
-0.03
0.03
SbVel
0.18
0.00
-0.14
0.07
-0.01
ScowDist
0.95
0.03
0.08
-0.05
-0.09
ScowV_D
0.91
0.07
0.10
-0.02
-0.09
ScowVel
0.91
-0.03
-0.15
0.02
-0.14
TotalBfill
0.10
0.94
-0.05
-0.01
0.02
VolRemTotal
0.87
-0.04
0.08
-0.04
0.12
tPCB CU AREA
0.49
-0.14
0.78
-0.23
0.01
BoatDist
0.85
0.18
0.08
-0.02
-0.02
BoatVel
0.61
0.07
0.02
0.05
0.00
BoatVD
0.83
0.14
0.12
-0.03
-0.01
LoadBoats
0.59
0.37
0.12
0.60
-0.12
T otalEfficiency
0.32
-0.09
0.16
0.91
0.06
ME
0.91
-0.10
0.10
0.23
-0.07
Semi-Partial R2
28%
6%
4%
5%
13%
Factor Interpretation
Volume/Mass
Bucket Fill
Flow Weighted
Backfill
PCB/Flow Weighted
CU Area
Flow Weighted
Vessel Dist.
Mass
Removed
Hudson River PCBs Site
EPA Phase 1 Evaluation Report - Addendum Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Table 2-4
Productivity and water column concentrations from 8/
10/2009 through 9/7/2009
Month
Day
Date
Productivity
(cy/day)
Water Column PCB (ng/1)
8
10
8/10/2009
151
8
11
8/11/2009
837
115
8
12
8/12/2009
1785
124
8
13
8/13/2009
2413
105
8
14
8/14/2009
2567
135
8
15
8/15/2009
2644
122
8
16
8/16/2009
97
8
17
8/17/2009
2773
144
8
18
8/18/2009
2421
166
8
19
8/19/2009
2212
198
8
20
8/20/2009
2212
135
8
21
8/21/2009
2260
223
8
22
8/22/2009
3559
176
8
23
8/23/2009
113
8
24
8/24/2009
3006
101
8
25
8/25/2009
3470
146
8
26
8/26/2009
3151
194
8
27
8/27/2009
3297
173
8
28
8/28/2009
3020
196
8
29
8/29/2009
2352
164
8
30
8/30/2009
256
8
31
8/31/2009
2796
171
9
1
9/1/2009
2231
189
9
2
9/2/2009
2466
187
9
3
9/3/2009
2450
214
9
4
9/4/2009
2325
206
9
5
9/5/2009
145
9
6
9/6/2009
72
9
7
9/7/2009
111
Average
2557
156
Hudson River PCBs Site The Louis Berger Group, Inc.
EPA Phase 1 Evaluation Report - Addendum Page 1 of 1 April 2010
-------
Explained Jointly
Mass Removed CU18 (Inside
Sheet Pile)
Tug Traffic for CU17-18
Tug Traffic for Backfilling
Flow at Ft. Edward
Mass Removal Efficiency
Scows in Que
Percent \feriance Explained
(127-day Mechanism Model)
o%
1
-
¦
-
]
10% 20%
Percent Variance (%)
30%
Note: Percent variance explained by individual process variables in a multiple
regression model predicting water column PCB concentration at Thompson Island Dam,
Hudson River New York. Overall adjusted R2=68%. Variance explained jointly cannot
be ascribed independently to any particular variable due to inter-correlations among the
predictors.
Percent Variance Explained by Individual Process Variables in a Multiple Regression
Model Predicting Water Column PCB Concentration at Thompson Island Dam
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 2-1
April 2010
-------
600
500
a
— 400
m
U
I 300
200
100
0
Modeled and Observed Total PCB
127-Day Mechanism Model With 166-Day Model Substitution
Including Mass RemovedCU18
r '
a
* 'Vt \ *
J\;
~a i*l t
* ~ k *..• *j
1-May
31-May
30-Jun
30-Jul
Day
29-Aug
28-Sep
28-Oct
- Modeled PCB ng/l —¦*—Observed PCB ng/l - - UPL95
Note: The model is based on process variables available on 127 ofthe 166 day season with modeled values from the 166 daymodel
substituted on the remaining days—primarily in May and June.
Observed and Modeled Values for Water Column PCB Concentrations at
Far-Field Station in Thompson Island Pool
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 2-2
April 2010
-------
Tug Traffic (CU-17/18 to Lock7)
(Distance-Depth-PCB Weighted)
in
CD
U
Q-
600
500
400
300
200
100
0
y = 90.658e
O.OOlGx
R2 = 0.3149
•
S
• ¦
•
%
•
•
•
•
•
• y
• /
• * X
- / •
•
•
• *
¦ •
V"
•
•
i i«*
r
•• •
n .
?• .
• •
•
••
•
• •
*
•
0
200
400 600 800
Mass Removed (kg)
1000
1200
Note: The line is the best fit line with an R2=69% and adjusted R2=68%
Observed Water Column Total PCB Concentration Plotted Against Modeled
Values for Thompson Island Pool Based on the 127 Day Model
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 2-3
April 2010
-------
Modeled and Observed Total PCB
127-Day Model With 166-day Substitution
600
500
400
3 300
o.
200
/.
/': V.
-\ . i i i a ~ ;;
t x-' - .-i'/Wt
f. --/v t.fn \m \ r *[. tk
trv \ \ 4m¦ m4 ^r¦
i I
1 V v W
~
oi
E
<13
¦i—I
o
100
0
1-May
31-May
30-Jun
30-Jul
29-Aug
Day
¦ Modeled PCB ng/l ------ Observed PCB ng/l UPL95
28-Sep
28-Oct
Note: The model is based on variables available on 127 of the 166 day season with modeled values from the 166 day model
substituted on the remaining days—primarily in May and June.
Observed and Modeled Values for Water Column PCB Concentrations at
Far-Field Station in Thompson Island Pool
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 2-4
April 2010
-------
FlowFt. Edward
Mass x Efficiency
Hq Traffic
7/8 - 8'9
8/10 - 9/7
7/8 - 8/9
8/10 - !i/7
7j8 - 8/9
8/10 - 9/7
Bask fill Tuc Traffic
Mass CU 18
ScDwQueuB
«
Hi
_Zj
15000
1D000
5000
80
80
i/.i
^ 40
20
4-
1
1
1
1
. M
¦-W-
1 1 1
i ¦
«
(D
2 ¦
3 -
9/11-10/5 10/8-10/27
7/8 - 8/9
8/10 - !i/7
7j8 - 8/9
8/10 - 9/7
Distribution of Six Key Process Variables During Selected Periods in 2009
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 2-5
April 2010
-------
TOPIC 3
REDISTRIBUTION OF CONTAMINANTS
DURING DREDGING
-------
Topic 3 - Redistribution of Contaminants during Dredging
The GE Remedial Action Monitoring Plan Quality Assurance Project Plan (RAM QAPP; May
2009) required that GE conduct a special study aimed at measuring the spatial extent,
concentration, and mass of Tri+ PCB-contaminated material resulting from dredging that
deposited in areas immediately downstream of dredging operations and may act as a potential
source of future contamination of the water column and downstream surface sediments (GE,
2009a).
The RAM QAPP specified that sediment traps be placed downstream of multiple Phase 1 dredge
areas representing different sediment characteristics and PCB concentrations expected to be
encountered during dredging operations. The following locations were defined in the RAM
QAPP for trap placement:
• Immediately downstream of CU-9, in the west channel at Rogers Island.
• Downstream of CU-16 in the Northern Thompson Island Pool (NTIP)
• Downstream of the dredging areas in the East Griffin Island Dredge Area (EGIA)
The study was performed downstream of the EGIA, with traps deployed in the configuration
presented in the QAPP downstream of CU-18 on July 9, 2009 and sampled approximately once
per week. This part of the study was completed on August 17, 2009 (GE, 2009b).
However the other planned locations were modified due to project logistics (GE, 2009c). In
accordance with the RAM QAPP, sediment traps were deployed downstream of CU-9 in the
west channel of Rogers Island on May 6-7, 2009. A total of 34 traps were installed at the
approximate locations shown in the RAM QAPP. The first round of data collection was
completed on May 14 (to provide background data prior to the start of dredging), with 32 of the
34 traps successfully re-located; however these samples were not submitted for analysis (GE,
2009b).
Because the background sediment trap samples were not analyzed, push cores were collected in
September 2009 in the vicinity of areas where sediment traps were deployed, and adjacent to
Sediment Sampling and Analysis Program (SSAP) core locations (Figures 1 and 2; GE 2009c).
The intent was to compare the concentrations in these baseline push cores to those observed in
the SSAP cores obtained nearby five years earlier. Originally, 27 cores and a number of co-
located cores were to be collected. However, because the new cores were potentially collected in
areas with less or no sediment deposition, only six cores were actually obtained, and no co-
located cores were among them (EPA, 2010). It should be noted that cores were never
envisioned as part of this analysis. If they had been, collection would have been required just
prior to the initiation of dredging to allow for use as a true baseline. EPA's concerns and
recommendations for this study were forwarded to GE (see Common Appendix of EPA Phase 1
Evaluation Report; EPA, 2010). While concentration increases were observed in five of the six
baseline push cores obtained, no radionuclide testing was performed to determine the age of the
sediments and confirm whether or not they actually were recently deposited. And, as discussed in
greater depth in the EPA Phase 1 Evaluation Report (see Section 1-4.3), due to the high degree of
variability in surface sediments, the observed increases do not provide a statistically significant
Hudson River PCBs Site Page 3-1
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
difference over the original observations (EPA, 2010). Further, the baseline push core locations
appear to be biased toward low-concentration areas, increasing the likelihood for follow-up
concentrations to be higher than the initial. A proper sampling design with an unbiased selection
of locations and adequate sampling size is required to evaluate whether or not surface sediment
concentrations outside of dredge CUs have increased (EPA, 2010). A more detailed discussion
and analysis of this aspect of the GE downstream target area contamination study is presented in
Chapter 1, Section 4.3 of the EPA March 2010 Phase 1 Evaluation Report.
With the initiation of dredging on May 15, 2009, considerable barge traffic began traversing the
river between CU-9 and downstream support areas, often passing through the area where the
traps were installed. Barge staging was also conducted in that area, requiring considerable tug
activity to move and maintain barge positions during May 16 high flows. The first round of post-
dredging sampling was attempted during the week of May 25, however only two intact traps
could be found (GE, 2009b). It is likely that the others were dislodged when a barge broke free
of its spuds and ran aground during a high flow event (GE, 2009c). No useable data was obtained
from this deployment of traps (GE, 2009b).
Rather than repeating the study downstream of CU-9, traps were installed in an area downstream
of CU-10 along the eastern shore of Rogers Island, with the number of units actually deployed
decreased to reduce the potential for disturbance by project-related vessel traffic. The RAM
QAPP also specified that study be performed downstream of CU-16. The study was not
implemented in this area due to project vessel mooring activities (GE, 2009b).
The GE sediment trap study did not address all the requirements described in the RAM QAPP,
and, as such, does not provide conclusive evidence that sediment re-distribution in the river was
caused by dredging. The presence of sediment in the traps prior to the initiation of dredging
suggests a dynamic system, one where everyday flow conditions result in the re-distribution of
sediments. And while sediments may be mobilized by dredging and transported to other areas,
the samples obtained on May 14 from the baseline deployment were not sent for analysis, so
there are no pre-dredging benchmarks for spatial extent, concentration, or Tri+ PCB mass for use
as a baseline measurement in evaluating trap samples obtained during and after dredging
activities. Sediments released due to dredging cannot be distinguished from sediments
undergoing transport as part of the natural state of the river, including normally occurring
recreational boat traffic and project vessel traffic. EPA's March 2010 report documents the
dynamics of sediment movement under natural conditions based on bathymetric changes in
sediment bed elevation (see Section 1-4.3).
Evidence that boat traffic can contribute to the mobilization of bottom sediments was observed
during collection of the transect study samples. Specifically, a sharp increase in the total
suspended solids concentration was measured in samples collected at NTIP transect station, and
GE indicated that the spurious TSS results were due to sediment disturbance by the sampling
vessel itself. It is important to note that the type of vessel used to collect the transect samples was
small, typical of what is on the Hudson during the summer and other recreational timeframes.
Figures 3-1 and 3-2 show the boat traffic in the vicinity of the sediment traps and push core
locations.
Hudson River PCBs Site Page 3-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Sediment trap PCB data have been used by GE to suggest that sediments with elevated PCB
concentrations (as high as 100 mg/kg) were being transported downstream due to dredging
operations, where they would continue to contribute PCBs to the water column and potentially
contaminate areas that were not delineated for dredging (GE, 2010). While there likely was some
re-distribution of bottom sediments due to the Phase 1 dredging, EPA does not believe it
occurred to the extent GE has stated.
Summary
The Hudson River is highly dynamic, resulting in sediment movement under normal conditions,
with recreational boat traffic and larger vessel movement during the open lock season causing
normal re-distribution of bottom sediments. Before dredging started particulate materials were
collected in sediment traps, but these were not analyzed. GE did not analyze baseline sediment
trap materials which might have shed additional light on the sediment trap data collected during
dredging. During dredging, tug traffic significantly impacted sediment resuspension, particularly
in the vicinity of these traps, and local resuspension may have resulted in over trapping of
particles.
References
EPA, 2010. EPA Phase 1 Evaluation Report. Prepared for the USEPA by the Louis Berger
Group, Inc. March 2010.
GE, 2009a. Remedial Action Monitoring Plan Quality Assurance Project Plan. Prepared for
General Electric Company, Albany, NY by Anchor QEA, LLC, Environmental
Standards, Inc., and Arcadis. May 2009.
GE, 2009b. Remedial Action Monitoring Program Corrective Action Memorandum No. 3.
Prepared
GE 2009c. Phase 1 Data Compilation. Prepared for General Electric Company, Albany, NY, by
Anchor QEA, LLC. November 2009.
GE, 2010. Phase 1 Evaluation Report. Prepared for General Electric Company, Albany, NY by
Anchor QEA, LLC, and Arcadis.
Hudson River PCBs Site Page 3-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
SI
^ o
o
o 1
O n «
CU-9
o
o o
T7
TT
ei.j 2.:J"
o
o
o
8.6
o o •
CU-10
)\o o o
1 ftll-3
Legend
Surface TPCB Concentration in SSAP Cores (ppm)
cCU
JO
•
o
o
CO
o
o
3.1 -10.0
o
10.1 -30.0
•
30.1 -100.0
o
100.1 -300.0
o
300.1 - 1000.0
•
1000.1 -1650.0
•
Missing TPCB Data
n
SSAP Core Location in the Vicinity
of Supplemental Sediment Cores
Sediment Trap Locaitons
200 ft Buffer of Sediment Trap Locations
Revised Certification Units
Boat Traffic during Sediment Trap Sampling Dates
within 200 ft of Sediment Trap Locations
I Z
Navigation Channel
0 50100 200
Feet
Shoreline (20081222)
2002 SS Sediment type
Silt
Silt and Sand
Gravel
Transitional
Bed Rock
Boat Traffic and Surface TPCB Concentration in SSAP and Push Cores
in the Vicinity of Sediment Trap Locations
EPA Phase 1 Evaluation Report - Hudson River PCBs Site
Figure 3-1
April 2010
-------
-------
TOPIC 4
FORECASTS OF DREDGING RELATED IMPACTS
AND RELATED RISK
-------
TOPIC 4-A
FORECAST OF PCB LOADS DUE TO DREDGING
COMPARED TO NATURAL ATTENUATION
-------
Topic 4-A - Forecast of PCB Loads Due to Dredging Compared to Natural Attenuation
Abstract
The analysis presented in this Topic revises the analysis set forth in Appendix I-G of EPA's
Phase 1 Evaluation Report, which stated that estimates of Total PCB (TPCB) loads at Waterford
based on the United States Geological Survey (USGS) and General Electric Company (GE) data
for the period 1995 to 2008 show no statistically significant decline, with "half life" of 99 years.
The analysis in Appendix I-G was revisited to improve the original calculation. In the reanalysis,
the analysis used PCB concentration instead of loads, since a better correlation could be
developed for concentrations to evaluate the recovery. Flows were simulated via a Monte Carlo
analysis rather than forecast as a single set of values. Additionally, further analysis of the
historical data suggested that the records for the years 1999 and 2000 were inconsistent with the
general trend observed in these data. Although there was no specific reason identified for the
inconsistency, the 1999 and 2000 data were excluded from the best estimate calculation of the
rate of concentration decline with time.
This Topic presents empirically-based projections of Monitored Natural Attenuation (MNA) of
Tri+ PCB load at Waterford, generated from historical data from USGS collected betweenl995-
2001) and baseline monitoring data collected by GE from 2004-2009. The results of this
analysis indicate a dredging release of approximately 500 kg of Tri+ PCB, as measured at the
Waterford far-field station, would provide net benefit to the river based on an overlap in the band
of forecasted uncertainty between this dredge release scenario and MNA. While the load
standard will continue to be based on a 1 percent net loss of TPCB inventory removed
(approximately 2000 kg TPCB or 667 kg Tri + PCBs), the goal will be to minimize the dredging-
related release of Tri+ PCB to 500 kg (as measured at Waterford) over the life of the project
(including both Phase 1 and Phase 2). Support for the adoption of control levels derived from
the 1 percent net loss of TPCB inventory removed is provided by consideration of the impact on
fish tissue concentrations and project risks in both the Upper and Lower Hudson River.
4-A.l. Introduction
One of the objectives of the Resuspension Performance Standard was to minimize the export of
PCBs to downstream areas, including the Lower Hudson (USEPA, 2004). This resuspension
export was defined as the net cumulative load of PCBs at Waterford. The Waterford station was
selected because it is downstream of the target areas identified in the feasibility study (FS)
(USEPA, 2000b) but upstream of the Mohawk River, which was shown to be a minor but
measurable source of PCBs to the Lower Hudson River (USEPA, 1997) with respect to
downstream impact.
During the development of the Resuspension Performance Standard, USEPA established
seasonal load standards for PCB export to the Lower Hudson River. As stated in EPA's Phase 1
Evaluation Report (USEPA, 2010), the estimated net loads due to dredging and related activities
at Waterford exceeded the standard set forth in the Resuspension Performance Standard,
although the load did not exceed 1 percent of the PCB mass removed, which was an important
factor underpinning the load criteria. EPA has specified several reasons why the load standards
as set forth in the Resuspension Performance Standard should be revised upwards (USEPA,
2010). These include, but are not limited to:
Hudson River PCBs Site Page 4-A-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
• The observed baseline loads to the Lower Hudson prior to dredging were substantially
greater than the previous model forecast of MNA;
• The surface sediment concentrations in the Upper Hudson River remain elevated despite the
passage of time and continue to provide a greater reservoir of contaminated sediments for
transport to the Lower Hudson than was envisioned when the remedy was selected; and,
• USEPA estimates that the inventory of PCBs in the Upper Hudson sediments may be at least
1.5 times the remedial design value of 115,000 kg TPCBs (Phase 2 Dredge Area Delineation
Report; GE, 2007) estimated prior to Phase 1 dredging activities.
The above reasons provide further impetus for the remedy, as well as for revising the threshold
for PCB load release to the Lower River in the Resuspension Standard. To establish a revised
load threshold, the long-term impact of dredging-related release relative to natural attenuation
needs to be assessed. To this end, this paper presents empirically-based projections of Tri+ PCB
load at Waterford that were generated for two scenarios:
1. The MNA Scenario - refers to the scenario which relies on natural recovery processes, within
the context of a carefully controlled and monitored site.
2. Dredging Impact Scenario - This scenario predicts the future water column export at
Waterford due to active remediation by dredging.
The empirical projections for both scenarios allow for uncertainty in future flows, based on the
past flow record. The methodologies used to develop projections for each scenario are presented
below.
4-A.2. Development of MNA Projections of Water Column Loads
The development of the MNA projection can be summarized as follows:
1. Review of the water column concentration data available for Waterford. The Tri+ PCB
data used to evaluate baseline conditions were obtained from the historical data (1995-2001)
from the USGS and the baseline monitoring program (BMP) data (2004-2008) collected by
GE. Review of the USGS historical water column concentrations at Waterford indicated that
the 1999 and 2000 values were mostly reported as non-detect. Data for these two years were
excluded from the baseline assessment, resulting in 172 Tri+ PCB data points remaining in
the USGS data set from 1995 to 2000. It is unclear why the 1999 and 2000 concentrations
measured by the USGS were non-detects. There are 339 Tri+ PCB data points in the GE data
set from 2004 to 2009. Table 4-A-l presents the statistical summary for the Tri+ PCB for
both data sets.
2. Estimation of Annual Baseline Loads. The annual Tri+ PCB loads were estimated by
individual year from 1995 to 2008. Although there are several load estimators, research has
shown that the Beale's Ratio Estimator is the only estimator that provides unbiased estimates,
although it requires stratification under event sampling (Preston et al., 1992). The Beale's
Ratio Estimator has been used widely in Great Lakes loading calculations, is thoroughly
discussed in Baun (1982), and was used previously for the Hudson River as part of the Data
Evaluation and Interpretation Report (DEIR, USEPA, 1997). In general, ratio estimators use
the period's data to calculate a mean daily load and then use the mean discharge from days
Hudson River PCBs Site Page 4-A-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
lacking concentration data to adjust the mean daily load (Richards, 1998). The AutoBeale
program (Richards, 1998) was used for the load calculations in this analysis. AutoBeale, a
computer implementation of Beale's Ratio Estimator, iteratively seeks out the discharge
stratification and minimizes the variance of the load estimate for a given set of data.
Stratification is the division of the sampling data into two or more parts which are different
from each other but relatively homogeneous within. It can be based on flow or time. Because
event sampling was conducted as part of the BMP at Waterford, stratification was necessary
to improve the precision and accuracy of the annual loads estimated.
3. Estimation of Flow-weighted Concentrations. For each year, the annual load estimated as
described above was normalized to the total flow volume for that year to determine the flow-
weighted average concentration.
4. Exponential Decay Fit to Flow-weighted Concentrations. The flow-weighted
concentrations were fit with an exponential decay function and this function was used to
forecast annual flow-weighted average concentrations.
5. Prediction of Annual Loads under MNA from 2009 to 2070. A spreadsheet-based model
was developed to estimate an empirical forecast of future Tri+ PCB loads at Waterford, and
to estimate the range of uncertainty in that forecast given the uncertainty in future flows.
Because there is additional uncertainty in the coefficients of empirical fit to the flow-
weighted concentrations, the simulated error bounds estimated in this analysis likely
understate the range of uncertainty in future loads. Estimated annual loads were predicted
from 2009 to 2070 (a 62-year forecast period). Annual flows from 1977 through 2008 at the
USGS Waterford gauge were randomly sampled using a Monte Carlo approach to represent
future flows. The annual loads for the 62-year period were simulated 1000 times, which
consisted of multiplying the forecasted concentration by a sequence of 62 randomly selected
calendar flow years from the period of record, 1977 through 2007. Thus, each year in the 62-
year period was represented by a complete calendar year of flow data, with each year being
selected at random, and a different sequence of years for each of the 1000 simulations. The
mean, 5th and 95th percentiles were estimated for the 62-year forecast annual loads.
4-A.3. Development of Dredging Impact Scenario
The development of the Dredging Impact projection can be summarized as follows:
1. Starting with the MNA projections, potential dredging-related loads (400 kg to 600 kg of
Tri+ PCBs) were applied to determine a dredge-related load that would result in a reduced
cumulative load to the Lower Hudson relative to MNA in the long-term. This improvement
was determined by the amount of overlap between the dredging scenario and the MNA
scenario within their respective uncertainty bounds. A load of about 500 kg due to dredging
release was established. This 500 kg dredging release was assumed to be transported
downstream to Waterford, and the load was evenly divided between the remediation time
period of 2009-2015, which represents a constant annual load of 72 kg/year above baseline.
This constant annual load was added to the baseline load as defined before dredging started,
consistent with the Resuspension Standard.
Hudson River PCBs Site Page 4-A-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
2. The reduction in load due to the dredging remedy was assumed to start in 2016, and this was
represented by an instantaneous best-estimate load decline of 60 percent relative to MNA in
that year. The 60 percent load decline was determined as follows:
a. The improvement in surface sediment concentrations was estimated relative to
pre-dredge values in the Thompson Island pool, the Schuylerville pool and the
Stillwater pool. The estimated declines in surface concentrations due to dredging
in these three sections of the river were: 87 percent, 30 percent and 6 percent,
respectively.
b. Using the GE BMP data from 2005 to 2008, the baseline load contributions from
these pools were reduced by the estimated improvement in surface sediments. The
total load reduction for Tri+ PCB varied from 45 to 80 percent, with a best
estimate of 60 percent.
3. Following the application of the instantaneous decline in 2016, the dredging scenario was
allowed to decay annually using the same rate as the MNA projections until 2070.
4-A.4. Findings
Table 4-A-2 presents the annual loads of Tri+ PCB estimated by the Auto Beale program. Figure
4-A-l presents the estimated annual weighted average concentrations and the exponential decay
fit. Note that 1999 and 2000 values are not included in the determination of the exponential fit.
Even after normalizing for flow, the concentration data appear like a step function and it is
unclear whether the decline estimated is due to a difference in sampling and analytical methods
between the USGS and the GE BMP programs.
Figures 4-A-2 and 4-A-3 show the annual project loads and the cumulative loads respectively for
the MNA scenario. Uncertainties are represented by the 5th and 95th percentiles of projected
loads in each year, given the simulated variability in the annual flows. For comparison,
HUDTOX forecast p3nas2, which represents the long-term Monitored Natural Attenuation
scenario used in development of the Resuspension Standard (USEPA, 2004), is also shown in
Figures 4-A-2 and 4-A-3.
Figures 4-A-4 and 4-A-5 show the annual and cumulative loads for MNA and the 500 kg
dredging release scenarios. Overlap between the uncertainty bands in future loads under the 500
kg dredging release scenario and the MNA scenario is projected to occur between 3 and 20 years
after dredging is completed. Furthermore, there is at least a 50 percent probability that the
cumulative load under the dredging scenario falls below the MNA cumulative load by the year
2050 and a 95 percent probability that the two scenarios cross during the 62-year simulation
period. While the load standard will continue to be based on a 1 percent net loss of TPCB
inventory removed (approximately 2000 kg TPCB or 667 kg Tri + PCBs), the goal will be to
minimize the dredging-related release of Tri+ PCB to 500 kg (as measured at Waterford) over
the life of the project (including both Phase 1 and Phase 2). Support for the adoption of control
levels derived from the 1 percent net loss of the TPCB inventory removed is provided by
consideration of the impact on fish tissue concentrations and project risks in both the Upper and
Lower Hudson River.
Hudson River PCBs Site Page 4-A-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
References
Baun, K. 1982. Alternative Methods of Estimating Pollutant Loads in Flowing Water. Tech.
Bulleting 133, Dept. Natural Resources, Madison Wisconsin. 11 Pages.
GE. 2007. "Hudson River PCBs Superfund Site Phase 2 Dredge Area Delineation Report."
Prepared by QEA for General Electric Company, Albany, NY. December 2007.
Preston, Stephen D., Victor J. Bierman Jr., and Stephen E. Silliman, 1992. Impact of Flow
Variability on Error in Estimation of Tributary Mass Loads." Journal of Environmental
Engineering 118(3): 402-419.
Richards, R.P. 1998. Estimation of Pollutant Loads in River and Streams: A guidance document
for NPS Programs. Project report prepared under Grant X998397-01-0, U.S.
Environmental Protection Agency, Region VIII, Denver. 108 p.
USEPA, 2010. Phase 1 Evaluation Report. Prepared for EPA Region 2 and USACE Kansas City
District by The Louis Berger Group, Inc. March 2010.
USEPA. 2004. "Engineering Performance Standard. Hudson River PCBs Superfund Site."
Prepared by Malcolm Pirnie, Inc. and TAMS Consultants, Inc. for USEPA. April 2004.
USEPA. 2000. "Hudson River PCBs Site Reassessment RI/FS Phase 3 Report: Feasibility
Study." Prepared for the US Environmental Protection Agency Region 2 and the US
Army Corps of Engineers Kansas City District by TAMS Consultants, Inc. December
2000.
USEPA, 1997. "Data Interpretation and Evaluation Report - Hudson River PCBs Reassessment
RI/FS, New York." United States Environmental Protection Agency. February 1997.
Hudson River PCBs Site Page 4-A-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 4-A-l
Summary Statistics of Tri+ PCB Concentrations at Waterford
Data Source
Month
Tri+ PCB (ng/L)
Count
Min
Max
1
7
10.0
89.0
2
4
5.0
10.0
3
21
1.0
67.3
4
39
2.2
116.0
5
28
5.0
78.0
USGS
6
10
5.0
50.9
7
7
5.0
40.0
8
7
5.0
37.0
9
13
2.5
46.0
10
16
10.0
70.0
11
12
5.0
114.6
12
8
5.0
54.8
USGS Summary
-
172
1.0
116.0
1
31
2.0
79.0
2
13
4.7
23.9
3
41
2.3
42.0
4
69
1.5
126.3
5
18
4.4
20.9
GE
6
29
11.2
153.6
7
21
9.9
45.1
8
23
4.2
28.6
9
22
7.6
18.5
10
23
7.5
36.6
11
36
3.6
82.0
12
13
1.9
9.3
GE Summary
-
339
1.5
153.6
Overall Summary
-
511
1.0
153.6
Hudson River PCBs Site Page 1 of 1
EPA Phase I Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 4-A-2
Estimated AutoBeale Annual Loads for Baseline Data
Year
AutoBeale
Annual Flow
Volume (billion
gallons/yr)
Weighted
Average
Concentration
(ng/L)
Load (kg/year)
Root Mean
Square Error
(RMSE) (kg/yr)
1995
164
19
1,477
29.4
1996
303
25
2,721
29.4
1997
226
11
1,894
31.6
1998
239
12
1,958
32.3
1999
38
0
1,535
6.6
2000
72
2
2,436
7.8
2001
163
13
1,609
26.7
2004
149
2
2,087
18.9
2005
133
7
2,372
14.8
2006
174
10
2,774
16.5
2007
103
6
2,063
13.2
2008
147
13
2,642
14.7
Hudson River PCBs Site Page 1 of 1
EPA Phase I Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
115
T3
a)
.? 10
a)
(0
3
C
C
<
y = 5E+61e-° 069x
R2 = 0.8935
0
1995
~i 1 1 1 1 1 1 1 r
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Year
Annual Weighted Average Tri+PCB Concentration vs. Year
at Waterford Station
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-A-l
April 2010
-------
140
120
Mean Tri+PCB for MNA Scenario
5%-tile Tri+PCB for MNA Scenario
95%-tile Tri+PCB for MNA Scenario
HUDTOX
2009
2018
2027
2036
2045
2054
2063
2070
Year
Simulated Annual Load for Tri+PCB
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-A-2
April 2010
-------
2,500
2,000
o>
g 1,500
m
o
CL
+
1,000
500
Mean Tri+PCB for MNA Scenario
5%-tileTri+PCB forMNA Scenario
95%-tile Tri+PCB forMNA Scenario
HUDTOX
0
2009
2018
2027
2036
2045
2054
2063
2070
Year
Simulated Cumulative Load for Tri+PCB
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-A-3
April 2010
-------
350
300
250 -
_l I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I L_
O)
^200
o
m
O
CL
F 150
100 -
50
1995
5th-95th percentile Tri+PCB for MNA Scenario
5th-95th percentile Tri+PCB for 500kg Dredging
Release Scenario
Regression Fit
Observed Load
2005
2015
2025
2035
2045
2055
2065 2070
Year
Simulated Annual Tri+PCB Load MNA and
500kg Dredging Release Scenario
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-A-4
April 2010
-------
3000
2500 -
_l I I I I I I I I I I I I I I I I I I I I I I I I I I I L
5th-95th percentile Tri+PCB for MNA Scenario
J I I I I I l I I I I I l I I l I I I I I I I l l I L_
2000 -
O)
T3
CO
O
_l
g 1500
CL
+
1000 -
500 "
2009
2019
2029
2039
Year
2049
2059
2069
Simulated Cumulative Tri+PCB Load MNA and
500kg Dredging Release Scenario
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-A-5
April 2010
-------
TOPIC 4-B
PREDICTION OF IMPACTS ON FISH TISSUE
CONCENTRATIONS IN THE LOWER HUDSON
RIVER DUE TO DREDGING
-------
Topic 4-B - The Impact of Dredging-Related PCB Release on Fish in the Lower Hudson
River
Introduction
The purpose of this analysis is to estimate the impact of dredging-related PCB release on fish
tissue concentrations in the Lower Hudson River for the expected duration of dredging (2009 and
2011 to 2015) and the brief {i.e., 5-8 year) period immediately beyond. The intention is to
examine any potential incremental increases in projected fish tissue PCB concentrations.
Additional risks (above current risk levels) to human health via fish consumption would be
identified if such incremental increases in fish tissue concentrations remain elevated above the
forecasted MNA recovery curve for the remedy selected in the ROD for many years after
completion of the dredging.
This analysis is based on a forecast of the mass delivered to the Lower Hudson over the
remediation period 2009-2015. As stated in the March 2010 EPA Phase 1 Evaluation Report,
between May 15 and November 30, 2009, the cumulative TPCB load to the Lower Hudson at
Waterford over baseline was about 150 kg/yr. The Tri+ PCB load over baseline for the same
period was about 60 kg. Although these loads are higher than were anticipated by the Phase 1
Resuspension Standard, they reflect the fact that the PCB mass removed from the actual Phase 1
area dredged was 1.5 to 1.8 times greater than the original estimate. Notably, the loads at Lock 5
and Waterford were close to or below the original criteria used in setting the Phase 1 standard,
that is, 1 percent of the mass removed.
Methodology
Based on the Phase 1 experience, EPA's current working estimate of the amount of PCB mass to
be removed from the sediments is 1.5 to 1.8 times the original estimated mass of 115,000 kg,
which was based on the design sediment sampling program. This results in a range of mass to be
removed of 170,000 kg to 210,000 kg. Although these estimates are still subject to review and
revision, they form a useful basis for exploring the potential impacts downstream in the Lower
Hudson.
As mentioned above, the Resuspension Standard was derived based on a mass loss or export loss
rate of 1 percent of the total inventory dredged in the Upper Hudson River. While the load
standard will continue to be based on a 1 percent net loss of PCB inventory removed
(approximately 2000 kg or 667 kg Tri + PCBs), the goal will be to minimize the dredging-related
release of Tri+ PCB to 500 kg (as measured at Waterford) over the life of the project (including
both Phase 1 and Phase 2). Support for the adoption of control levels derived from the 1 percent
net loss figure is provided by consideration of the impact on fish tissue concentrations and
project risks in both the Upper and Lower Hudson River. To examine the impact of these loads
on the lower river, the models used to assess PCB fate and transport and biological uptake in the
Lower Hudson for the purposes of the ROD and the Performance Standards were applied to
forecast fish tissue concentrations. Two net loading scenarios at Waterford were examined; they
correspond to dredging-related TPCB release loads of 1,800 and 2,400 kg, with corresponding
Tri+PCB release loads of 600 and 800, kg. These values were selected prior to the finalization of
Hudson River PCBs Site Page 4-B-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
the proposed revision to the load standard and serve as upper bound estimates on the impacts of
the 500 kg Tri +PCB scenario. Tables 4-B-la and 4-B-lb list the dredging-related TPCB and
Tri+PCB release loads respectively for these scenarios. It should be noted that here the models
for both the Upper and Lower Hudson simulate Tri+PCBs. The estimates for Total PCB
transport are then obtained by application of the appropriate Total PCB to Tri+ PCB ratio.1
The water column and sediment concentrations in the Lower Hudson River were estimated from
the same model of chemical fate and transport that was used in the Performance Standard (Farley
et al. (1999)), referred to herein as the Farley model. The Farley model was run for the two
dredging-related PCB release scenarios to estimate sediment and water column concentrations of
the Lower Hudson out to 2046. The fish tissue concentration was predicted by interpolating
previous results from a bioaccumulation model developed for the Hudson River remedial
investigation and used in both the ROD and the Performance Standards (FISHRAND, USEPA,
2000).2
Figures 4-B-la and 4-B-lb show the composite fish tissue concentration for the two scenarios for
dredging-related release scenarios of PCBs, along with the previously run Monitored Natural
Attenuation (MNA) scenario in the Lower Hudson River (for RMs 152, 113, 90 and 50). While
preparing these scenarios, the revised forecast of the post-dredging reduction in the Upper
Hudson load due to dredging and source control was not yet available. As a result, these
forecasts were assessed based on the return to the MNA recovery curve, and do not account for
the improvements to Lower Hudson conditions resulting from the load reduction from the
remediation and source control efforts. Thus these curves should represents a maximum time to
rejoin the MNA recovery curve since the reduction in the load from the Upper Hudson would
cause a faster recovery of fish tissue than simply returning to the baseline load conditions.
For these forecasts, the fish tissue concentration asymptotically returns to the MNA curve.
Notably in all instances, the curve returns within a limited amount of time after 2015, the
simulated year of completion of the remedial operation. If the decrease in load to the Lower
Hudson due to the remedy was taken into account, the fish tissue concentrations after
remediation would decline to a level lower than that of the MNA, as shown in Figures 4-B-2a
and 4-B-2b, taken from the original Performance Standard report. In these analyses, fish tissue
concentrations are estimated while accounting for the decline in PCB loads resulting from the
remedy. The difference in dredging-based load is relatively inconsequential since both high and
low load estimates decline rapidly after completion of the remedy. Recall that the purpose of the
present analysis was to demonstrate the impact of dredging-related PCB release on fish tissue
concentrations in the Lower Hudson River for the expected duration of dredging (2009 and 2011
to 2015) and the brief {i.e., 5-10 year) period immediately beyond.
1 The Total PCB to Tri+ PCB ratio varies with several factors. Sediments commonly have a ratio of 3:1 whereas
baseline water column PCBs range from 1.7 to 2.3, depending on location and time of year.
2 The FISHRAND model essentially relates fish tissue to water column and sediment concentrations on a linear
basis, so that the interpolation from prior runs to the specific conditions simulated for this analysis by the Farley
model is a straightforward exercise. Complete runs of the FISHRAND model are currently underway and could not
be completed in time for this presentation. However, given the precision of the interpolation, model results are not
expected to differ from the interpolated results by more than 1 or 2 percent.
Hudson River PCBs Site Page 4-B-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
It can be seen from the plots in Figure 4-B-la that for RM 152, the fish tissue concentration
asymptotically returns to the MNA curve about 5 years after dredging is completed for both
scenarios. This is essentially the same time to return to the no-resuspension recovery curve as
predicted for smaller loads in the original Farley model-FISHRAND runs (see Figure 4-B-2)3' 4
Similar to what was observed in the analyses for the ROD, the fish concentrations take a few
years longer to return to the target MNA recovery curve for the lower river miles. For each
scenario, Table 4-B-2 shows the number of years to return to the MNA recovery curve, an
approximation of the return to the no resuspension recovery curve. The time to return to the
MNA recovery curve in each case is similar to the time to rejoin the MNA recovery curve
originally forecast for smaller dredging releases in the analysis for the ROD and is generally no
more than 2 to 3 years longer. Additionally, the locations with the longest times to rejoin the
MNA recovery curve have the smallest fish tissue concentration spike from the curve. For
example, the 2,400 kg Total PCBs release simulation causes a maximum increase in the fish
tissue concentration of about 0.3 mg/kg at RM 50 as opposed to a shorter lived increase of about
1 mg/kg at RM 152. These model predictions lead to the following important conclusion: since
these scenarios all return to the MNA recovery curve relatively quickly, the larger releases
examined here are not expected to impact the time to reach the remedial goal for fish tissue of
0.05 mg/kg. They are also predicted to have little, or minimal, impact (2 to 3 years) on the time
to various interim remedial target levels5. Thus, there are no predicted impacts on the goals of the
ROD. Further, as noted above, these model predictions do not account for the improvements to
Lower Hudson conditions that will result from the load reduction from the remediation.
Summary and Observations
The Farley-FISHRAND models were run for a series of Upper Hudson loads that correspond to
the upper end of the proposed revisions to the Resuspension Standard. In particular, scenarios
were run for dredging-related loads of 600 kg and 800 kg of Tri+ PCBs (approximately 1800 and
2400 kg Total PCBs) at Waterford. The impacts of these loads were contrasted with the MNA
recovery curve, used here as a surrogate for the remedy curve response in the Lower Hudson.
This approach provided an upper bound on the likely time to rejoin the ideal recovery curve
(representing no resuspension load due to dredging).
The results show only minimal increases in the time to rejoin the ideal recovery curve relative to
scenarios representing less PCB loss to the Lower Hudson. From this analysis, it can be
concluded that the larger loads examined here have negligible long term impacts to Lower
3 The no resuspension curve shown in Figure 4-B-2 represents the ideal remediation curve, with no dredging-related
resuspension. Attainment of that curve a short time after dredging indicates that the impacts of dredging are short-
lived and have little influence on the improved rate of recovery anticipated by the remedy.
4 The original Farley model-FISHRAND runs assumed the remedy would be conducted between 2006 and 2011.
Therefore, the end of the remediation for those scenarios is the year 2011. Currently, it is anticipated that the remedy
will be completed in 2015.
5 The potential impacts due to dredging resuspension were recognized as part of the preparation of the ROD and
were included in the Responsiveness Summary. The impacts predicted here are similar to those noted in the
supporting ROD documentation. See for instance, the times to recovery shown in Table 363176-5 of the
Responsiveness Summary (USEPA, 2002).
Hudson River PCBs Site Page 4-B-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Hudson fish tissue concentrations and indicates that similar to the original load standard, the
revised load standard serves a dual role of protecting the Lower Hudson while ensuring that
dredging operations do not cause unnecessary resuspension.
In support of these model findings, there are several observations from Phase 1 that would
suggest this analysis is actually an upper bound on the impacts to the Lower Hudson. Despite the
increase in PCB loads measured during Phase 1 at Waterford, the monitoring of water column
concentrations in the Lower Hudson did not show any increase in the Tri+ PCB concentrations at
Albany or in Total PCBs and Tri+ PCBs concentrations at Poughkeepsie and the other Mid-
Hudson monitoring locations. The fish tissue results for the fall 2009-collected pumpkinseed and
forage fish, on a river section basis, in River Section 3 (approx. 168.2-183.2) of the Upper
Hudson did not show any increase of the mean value from the prior 5 years monitoring.
Appendix I-C of the EPA Phase 1 Evaluation Report (March 2010) discussed the resident fish
annual monitoring data in details.
Taken together, these observations and calculations indicate that the proposed range of revisions
to the allowable load standard for the Lower Hudson would not have any measurable long term
impacts to the recovery of fish tissue concentrations and therefore would not pose any additional
long term risks to Hudson River fish-consuming populations in the Lower Hudson.
References
Farley, K.J., R.V. Thomann, T.F. Cooney, D.R. Damiani, and J.R. Wand., 1999. An Integrated
Model of Organic Chemical Fate and Bioaccumulation in the Hudson River Estuary.
Prepared for the Hudson River Foundation. Manhattan College, Riverdale, NY.
USEPA, 2000. Phase 2 Report: Revised Baseline Modeling Report, Hudson River PCBs
Reassessment RI/FS. Prepared for EPA Region 2 and the US Army Corps of Engineers
(USACE), Kansas City District by TAMS Consultants, Inc. January 2000.
USEPA, 2002. Responsiveness Summary for the Record of Decision. Prepared for EPA Region
2 and the US Army Corps of Engineers (USACE), Kansas City District by TAMS
Consultants, Inc. January 2002.
Hudson River PCBs Site Page 4-B-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 4-B-la. Scenarios for Dredging Related Release Tri+ PCB Load at Waterford
Number of
Dredging Related Tri+
Dredging Related Tri+
Year
Start Date
End Date
Dredging
PCB Load for the
PCB Load per Day Net
Days
600 kg
800 kg
600 kg
800 kg
2009
5/16/2009
10/27/2009
134
59
59
0.44
0.44
2011
5/15/2011
10/1/2011
114
108
148
0.95
1.30
2012
5/15/2012
10/1/2012
114
108
148
0.95
1.30
2013
5/15/2013
10/1/2013
114
108
148
0.95
1.30
2014
5/15/2014
10/1/2014
114
108
148
0.95
1.30
2015
5/15/2015
10/1/2015
114
108
148
0.95
1.30
Total Release
600
800
Table 4-B-lb. Scenarios for Dredging Related Release Total PCB Load at Waterford
Year
Start Date
End Date
Number of
Dredging
Days
Dredging Related Total
PCB Load for the
1,800 kg
2,400 kg
Dredging Related Total
PCB Load per Day Net
1,800 kg
2,400 kg
2009
2011
2012
2013
2014
2015
5/16/2009
5/15/2011
5/15/2012
5/15/2013
5/15/2014
5/15/2015
10/27/2009
10/1/2011
10/1/2012
10/1/2013
10/1/2014
10/1/2015
134
114
114
114
114
114
176
325
325
325
325
325
176
445
445
445
445
445
1.31
2.85
2.85
2.85
2.85
2.85
1.31
3.90
3.90
3.90
3.90
3.90
Total Release
1,800
2,400
Note: The total PCB load was derived by multipling Tri+PCB load by 3, reflecting the Total to Tri+PCB ratio commonly found in the
sediments, which are the source of these loads.
Hudson River PCBs Site
EPA Phase 1 Evaluation Report Addendum
Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
Table 4-B-2
Year and Number of Years Composite Fish Tissue Concentration Recovery
Scenario
Number o
After C
RM 152
f Years for
Dmpletion c
RM 113
Fish Tissue ]
f Dredging (
RM 90
Recovery
2015)1
RM 50
350 ng/L2'3
5
6
9
11
Additional 1,800 kg TPCB
5
7
10
12
Additional 2,400 kg Total PCB
5
8
11
13
Notes:
lrThe fish tissue concentration was considered to have recovered when the forcasted fish tissue concentration for remedial scenario
agreed with MNA forcast to within 10% of the MNA value.
The 350 ng/L was previously run for the Performance Standard. The figure is included as Figure 4-2 for this addendum.
For the 350 ng/L scenario, the end of dredging was 2011.
Hudson River PCBs Site
EPA Phase 1 Evaluation Report - Addendum
Page 1 of 1
The Louis Berger Group, Inc.
April 2010
-------
#J3
~£fc
S
PS
u
a.
+
•c
H
s
ei
a>
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Species-Weighted Fish Body Burdens- Lower River RM 152 (1)
• MNA Forecast
¦Additional 1,800 kg Total
PCB (600 kg Tri+f
Additional 2,400 kg Total
PCB (800 kg Tri+f
2005
2010
2015
2020
2025
Year
2030
2035
2040
2045
1 Species-Weighted Average assumes 47% largemouth bass + 44% brown bullhead + 9% Yellow Perch
Composite Fish Tissue Concentration for the Lower River
1,500 to 3,000 kg Total PCB Release Scenarios
EPA Phase 1 Evaluation Report -Addendum - Hudson River PCBs Site
Figure 4-B-la
April 2010
-------
2005
Species-Weighted Fish Body Burdens- Lower River RM 90 (1)
»MNA Forecast
»Additional 1,800 kg Total PCB
(600 kg Tri+)
Additional 2,400 kg Total PCB
(800 kg Tri+)
2010
2015
2020
2025
Year
2030
2035
2040
2045
2.0
1.8
1.6
M
1.4
~£fc
S
1.2
PS
u
a.
1.0
+
'•I
H
0.8
s
ei
a>
2
0.6
0.4
0.2
0.0
2005
Species-Weighted Fish Body Burdens- Lower River RM 50 (1)
2010
2015
2020
2025
Year
2030
2035
2040
2045
1 Species-Weighted Average assumes 47% largemouth bass + 44% brown bullhead + 9% Yellow Perch
Composite Fish Tissue Concentration for the Lower River
1,500 to 3,000 kg Total PCB Release Scenarios
EPA Phase 1 Evaluation Report -Addendum - Hudson River PCBs Site
Figure 4-B-lb
April 2010
-------
81
DJj
s
cs
u
a.
+
c
H
w
2005
Composite Fish - RM 152
MNA (p3nas2)
No Resuspension (d004)
Total PCB 350 ng/L (sr04)
Total PCB 600 g/day (siOl)
2010
2015
2020
2025
Year
2030
2035
2040
2045
1 Species-Weighted Average assumes 47% largeinouth bass + 44% brown bullhead + 9% Yellow Perch
Composite Fish Tissue Concentration for the Lower River
Figure 2-9 of the Resuspension Performance Standard
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-B-2a
April 2010
-------
1.6
1.4
2_
+
ij
tr.
E 1.0
0.8
s 0.6
0.4
0.2
0.0
2005
Composite Fish - RM 90
2010
2015
2020
2025
Year
2030
2035
2040
2045
1.6 1
1.4
1.2
sx
E 10
U
± °'8
'mm
H
§ 06
o
o.o H
Composite Fish - RM 50
J 1 1 l—W 1 1 1_| . 1 1 1 1 1 1 . 1 1 L
2005 2010 2015 2020 2025
Year
2030
2035
2040
2045
1 Species-Weighted Average assumes 47% largemouth bass + 44% brown bullhead + 9% Yellow Perch
Composite Fish Tissue Concentration for the Lower River
Figure 2-9 of the Resuspension Performance Standard
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 4-B-2b
April 2010
-------
TOPIC 5
SCOW UNAVAILABILITY AND ITS
IMPACT ON PRODUCTIVITY
-------
Topic 5 - Scow Unavailability and Its Impact on Productivity: An Assessment of Scow
Queuing Effects
Abstract
The effects of scow load thickness and unloading time were evaluated both through examining
the scow unloading dock records from Phase 1 and through developing a probabilistic model
based on the unloading dock records (taken from Appendix P of GE's November 2009 Phase 1
Data Compilation Report). The number of scows that were at the loading dock (on queue and
being serviced) was calculated from these records. Although these records do not account for all
scows that were on queue (when the loading dock was close to full, some scows were moored in
the river and were not accounted for as being on queue), they show that productivity declined
when queues reached peak levels. It is also possible to discern that up to 14 of 18 scows were
unavailable in the week preceding the August 7th resuspension shut-down and that most scows
were available for Phase l's peak one-day productivity on September 8th.
The probabilistic model randomly pulled scow arrival and unloading data from Appendix P to
simulate the average conditions during Phase 1. There was close agreement between the model
and the actual average conditions, so the model was used to assess two factors: loading the scows
with more sediment on average and adding a second unloading station. Both of these factors
relate to unloading efficiency. The model shows that a single unloading station cannot meet the
Phase 2 monthly productivity target of 96,000 CY for a sustained period of time, even if scows
carry greater loads on average. This is because the queue that develops from the single
unloading station causes productivity to fall off, even when using only data from the best 30-day
period of Phasel. However, the model shows that the target can be met by adding a second
unloading station and increasing the scow loads by 0.7 feet above the average scow loading
thickness from Phase 1. This increase of loading is within the operational range of the scows and
should be achievable with the equipment on hand.
5.1. Introduction
During Phase 1 a single unloading station was utilized to unload sediment-laden scows. Notably,
the scow unloading service capacity was exceeded by the number of scows arriving at the
loading dock throughout much of the project. This had a downstream effect of limiting the
number of scows available for dredgers to fill during dredging. As discussed in EPA's Phase 1
Evaluation Report, about 26 percent of the available dredging time was lost to dredgers waiting
for scows. The unloading system used during Phase 1 was evaluated to assess how significantly
the unloading system affected scow availability and productivity. The system was ultimately
simulated with a probabilistic model of scow queuing based on the Phase 1 data to see if
modifications (either to the loading of scows or adding a second unloading station) would result
in higher productivity.
5.2. Development of Model Variables
Information on the arrival time, service start time, service end time, sediment load (in cubic
yards), and water removed (in gallons) for each scow unloaded during Phase 1 is provided by GE
in Appendix P of their November 2009 Phase 1 Data Compilation Report. The unloading dock
can accommodate up to 5 to 6 scows in a queue with a sixth or seventh scow being serviced by
Hudson River PCBs Site Page 5-1
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
the unloading station. However, scow logs show that many of the scows spent at least some
time at a river mooring before arriving at the unloading dock. This information is in handwritten
logs and was not compiled as part of Appendix P, while the records in Appendix P only account
for scows that are tied up at the unloading dock waiting to be unloaded and being unloaded. So
when the records show there are four or more scows waiting at the loading dock behind the scow
being serviced, it is likely that, due to limited space, scows were also being staged at moorings in
the river and a complete count of idle scows was not being made. That is, when 2/3 of the room
at the dock was occupied, it may have been easier to moor a scow in the river than it was to
maneuver scows already tied up at the loading dock to accommodate the new scow, even if there
was technically enough room for it. This is important for interpreting model results because the
full length of the queues during Phase 1 dredging often cannot be known for certain since the
number of scows moored in the river at any given time was not recorded and they could have
been moored there even at times when there is theoretically space available at the unloading
dock. Figure 5-1 shows the number of scows queued at the unloading dock while each scow was
being serviced.
Also shown on Figure 5-1 are GE's estimate of daily dredged volume and the daily estimated
volume arriving at the unloading dock. The peak queue occurred on July 24th with six scows in
line at the unloading dock. This period of scow unavailability precipitated a series of events
related to productivity: dredging productivity, which reached a peak on July 29th, subsequently
dropped and daily loads arriving at the unloading dock declined as well. Productivity (measured
both by GE's estimate of dredging and the amount of material arriving at the unloading dock)
had already begun to decrease due to scow unavailability in late July, a full week before the
August 7th resuspension event. The volume of material arriving each day also dropped six days
before the resuspension-related shut down. These events occurred during the period when GE
was attempting to meet the full-scale, maximum monthly productivity threshold of 89,000 CY
and are germane to assertions in GE's Phase 1 Evaluation Report that productivity thresholds
cannot be achieved without jeopardizing the resuspension standard water quality threshold. One
very important relationship that can be seen from Figure 5-1 is that scow queuing, and not
resuspension-related operational interruption, was controlling productivity. Further, since
productivity was already declining in the week leading up to the resuspension-related shut-down,
that resuspension event cannot have been precipitated by increasing production.1
Scows continued arriving at the unloading dock for three days after August 7th although no
dredging was taking place. The loads in these scows are not significantly different (less) than
scows arriving before or after. Four scows were in the queue at the start of August 8, and 10
scows arrived between August 8 and August 11th. This means that 14 of 18 scows were full and
ready to be unloaded on August 7th, with dredging occurring in all 10 CUs at that time. From this
it can be surmised that a maximum of only 4 out of 18 scows were available to service the 11
dredges that were working in 10 separate CUs on August 7th. This shutdown period highlights
how large the scow queue had become and how the scow queue was affecting productivity. The
1 This assessment raises the question as to what factors, if not an increase in productivity per se. may have
contributed to the exceedance of the 500 ng/1 water quality threshold that led to the operational shut-down on
August 7th. This question is addressed in separate papers in this Addendum under Topics 2 and 8 as is the validity of
the sample results from Thompson Island station autosampler during that time frame.
Hudson River PCBs Site Page 5-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
greatest daily dredging volume estimates occurred September 8th and 9th following 4 days
without dredging (i.e., the Labor Day weekend), and precede the second resuspension-related
event on September 11th. The likely explanation for this dramatic increase in productivity is that
for the first (and only) time since the start of dredging in earnest, nearly all scows were available
for use (except after the August 7th shutdown, when dredging was re-started slowly). Our
analysis in Topic 2 shows that the responsibility for the September 11th resuspension event can
be linked to the releases emanating from the windows in the sheet pile at CU18.2 Within two
days of resuming dredging after the long weekend (by September 9th), the queue had returned to
2 to 3 scows in line and production dropped by 700 to 900 CY per day.
A model was developed to assess two factors that may have contributed to generating and
maintaining the scow queue. The first factor is the limitation of the scow unloading station and
the second factor is the removable volume of sediment that arrived in each scow. Appendix P
provides information on both factors. The unloading start times and end times for each scow are
presented for each scow (however the time required to move the scow into place, the scow
handling time, is not documented specifically). The unloading times from Appendix P were used
as a basis for the service time in the model along with a calculation of the handling time
(described below). Based on the average large hopper scow dimensions, an estimate of load-
thickness to volume of sediment was calculated. Due to unloading equipment limitations,
between six inches and a foot of sediment was left in the scow after unloading, so the volume
unloaded was the material above this level, referred to here as the removable portion of the scow
load. This removable load thickness averaged about 1.7 feet during Phase 1. Figure 5-2 is a
histogram of scow removable load thickness from Phase 1. The removable load thickness is the
volume unloaded from each scow as given in Appendix P divided by the average scow bottom
area. It was noted during field activities that the last foot of material unloaded from a scow
required the greatest amount of time, so a relationship between scow removable load thickness
and scow unloading time per foot was evaluated. Figure 5-3 shows the relationship between
removable load thickness and the per-foot unloading time. This figure uses the average time for
all scows of similar removable load thickness. As can be seen, the more fully loaded a scow
was, the faster, on average, each foot of material was unloaded. This rate increase is due to the
rise in bucket efficiency of the excavator unloading the scow as the thickness of material in the
scow increases.
Scow handling time, the time between when an unloaded scow leaves the dock and the next
scow from the queue is available to be unloaded, was not specifically accounted for in Appendix
P. Scow handling time is part of the service time and represents the time necessary to move
scows from the queue to the unloading station. Appendix P only includes the time when a scow
was being unloaded. However, handling time could be calculated as the duration between scow
service events when scows were queued or, if no scows were waiting in a queue for service, as
the elapsed time from the arrival of a scow to the time that scow was serviced. Figure 5-4 is a
histogram of the scow handling time calculated from Phase 1.
2 Although productivity was at its highest levels in the days preceding the September 11, analyses show the direct
cause for higher PCB levels at Thompson Island on September 11 is releases that occurred from the silt curtain
enclosure at CU-18 as discussed in the paper on causes of resuspension in this Addendum under Topic 2.
Hudson River PCBs Site Page 5-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
5.3. Model Calibration and Baseline Application
A model of the scow queuing and unloading process was built in Microsoft Excel based on a
probabilistic framework, using the modeling procedures described in Winston (2004) and
Winston and Albright (2009). The model uses both Monte Carlo and bootstrap approaches to
access data from Phase 1 as input. The methods for each data set used are discussed below. The
model simulates scows arriving at the unloading dock (arrival interval and removable load
thickness) and being serviced (handling time plus service time, as calculated from the
exponential relationship with removable load thickness). Two 'clocks' are used: the arrival
clock determines when a scow arrives and the service clock determines when a scow is serviced
(both are relative to the start of the simulation). When a scow arrives it is either serviced or, if
there is already a scow being serviced, it is put in the queue. The ultimate number of scows and
the estimated loads of each scow serviced in a 30-day period were summed as output from the
model.
The model is based upon the following equations and conditional statements:
Queue length: Q (number of scows in queue)
Arrival clock: C \ (cumulative hours since simulation start)
Arrival interval: A (time between scow arrivals, in hours)
ft
For flit arrfifgi ef ffci ntksQQWi ¦ Y .4.
Service Clock: Cy (hours of scow unloading since simulation start)
Unloading time: U (hours)
Handling time: H (hours)
i a
iw the + St'}
where the unloading time is given by one of two formulas:
based on the average unloading time in Phase 1
Uem = 4.313S«-C"«*
based on the average unloading time in the 30 higher production days
= 3,3 43Stf-c
and b is the Removable Load Thickness in feet
b = (scow load volume removed (CY)/scow pay load area (Y2))*3 ft/Y
For each scow arrival:
If CA(n) < Cs(n-i) the scow is added to Q; Q = number of scows where Caw k Cs(n-i)
and Cs(n) is then the future time at which the nth scow will be serviced.
If CA(n)> Cs(n-i) the scow is serviced immediately; 0 0
When CA(n)> Cs(n-i) then Cs(n) is set equal to Caw since Cs(n) cannot be less than the scow
arrival time.
Hudson River PCBs Site Page 5-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Volume:
For Cs(n) = 720 hrs (30 days)
Scow count: S = number of scows that arrived in 30 days
The unloaded volume for each scow (in CY):
Vu = b*scow pay load area*CY/27ft3
The total unloaded volume (in CY):
TV = X^nsodays (the sum of sediment unloaded in 30-days)
Single model run yields: TV, S, 30-day average Q, 30-day average b
Monte Carlo output (750 model runs): Average TV, Average S, Average Q, Average b,
minimum TV., maximum TV, TV standard deviation.
The model baseline application follows this progression:
1. The Phase 1 data are arranged in a table with each row representing a discrete scow arrival
record, and the columns representing the attributes recorded for each arrival (load thickness,
arrival time at the unloading dock, etc.)
2. Two random numbers are generated; the first is used to select a scow arrival interval
randomly from the corresponding row in the data table, and the second to select a scow
removable load thickness randomly from its corresponding row (which, except for an
unusual coincidence, would be different than the row from which the scow arrival interval
was randomly selected).
3. The handling time is selected from the handling time distribution using the first random
number (that is, from the same row/record as the scow arrival interval).
4. The unloading time is calculated for the arriving scow using the exponential relationship
between removable load thickness and service time (see Figure 5-3) and added to the
handling time to determine the total elapsed service time.
5. The arrival interval is added to the arrival clock and the total service time is added to the
service time clock.
6. The arrival clock for the current scow is compared to the service clock for the completion of
the previous scow; if the scow has arrived before the previous scow's service is complete, the
scow is added to the queue.
7. The model then returns to step 1 to simulate the next scow arrival.
8. The total numbers of scows arriving for a 30-day period based on the service clock are
summed, as well as the total sediment volumes.3
3 To maintain simplicity, rather than directly programming a simulation limit on the number of scows to the project
total of 18, the model tracks the number of scows that have arrived. If the count in the queue exceeds 18 (including
the scow being serviced), these excess scows and their associated volumes are excluded from subsequent
calculations since they represent scow arrivals that could not occur.
Hudson River PCBs Site Page 5-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
9. An Excel "macro" was written to capture the iterated output for 750 model runs. The macro
re-randomizes Excel after every 50 runs to assure no repetition of data selection patterns
occurs.
10. The number of scows, total volume, average number of scows on queue, and average
removable load thickness are reported for the 750 runs along with minimum, maximum and
standard deviation for the total volume unloaded.
The model includes the use of the Excel random number generator to choose scow thicknesses
and arrival intervals (the time between two arriving scows) from the 628 scow arrivals that
occurred during Phase 1. With 18 scows and up to 11 dredges operating in up to 10 CUs, the
logistics of the project were complex and arrival interval and scow load thickness were shown to
be statistically independent variables with R2 <0.01 (see Figure 5-5). For this reason, separately
generated random numbers are used to select scow arrivals and scow removable load thicknesses
from the Phase 1 data pool, rather than selecting these factors from the data for the same arriving
scow with a single random number.
Scow handling time is selected probabilistically from the distribution shown on Figure 5-4 using
the random number generator as well. The scow service time is calculated using the removable
load thickness relationship shown on Figure 5-3 and added to the scow handling time to produce
a total service time. The total service time for each scow is then cumulatively summed to create
the service time clock to track how much time has passed since the first scow arrived and began
being serviced. The times between arrivals are also summed to create the arrival time clock
tracking the time associated with scow arrivals simulated to have passed since the first scow
arrived at the loading dock and began being serviced. The number of arrivals simulated to have
occurred between the time an individual scow arrived and completion of that scow's unloading
are counted as being in the queue.
The total amount of sediment unloaded in a 30-day period is then summed from the scows
simulated to have been serviced during that period according to the service clock. The average
scow removable load thickness is also calculated for the 30-day period, as well as the average
scow queue length. The model was run 750 times as a Monte Carlo simulation (with the input
variables drawn randomly for each run as described above) and the values for each of the output
parameters were averaged from the 750 runs to determine the average model output. Table 5-1
compares the average model output with the actual 30-day averages from Phase 1.
Table 5-1 - Model Output and Phase 1 Average Values
Average 30-day
Volume
Number of
Scows Serviced
Average
Number of
Scows in Queue
Average Scow
Removable
Load Thickness
Phase 1
47,338 CY
112.7
2.7*
1.67 ft
Model
47,331 CY
111.8
4.5
1.67 ft
* The 'average number of scows in queue' value for Phase 1 only includes scows that were docked at the unloading
dock. In reality, when the loading dock was near full, the queue continued to moorings in the river. During Phase 1,
the count at the dock was 4 or more scows on 139 occasions, and many scow logs include time at river moorings
before being moved to the unloading dock. For this reason the close agreement of the other factors suggests that the
Phase 1 average number of scows in the queue is an underestimate, rather than the model yielding an overestimate.
Hudson River PCBs Site Page 5-6
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
5.4. Model Predictions for Phase 2
Given the close agreement between the model output and the data for the actual 2009 conditions,
the model was then used to predict the impact on dredging production if additional material were
added to every scow and if an additional unloading station were added. For this evaluation, the
best 30 consecutive days of scow unloading were used for random values of removable load
thickness, arrival time and scow handling time.
The following modifications were made to the model to facilitate these analyses:
1. A term was added to the Phase 1 scow removable load thickness that would increase the
removable load thickness of each scow from Phase 1 by a specified amount (these are the
load thicknesses selected randomly by the model).
2. A flag was added to indicate that two unloading stations were active (the model can run
either one or two stations).
3. A logical statement was added to simulate two stations being active so that if one station is
occupied (as determined by that station's service clock) then an arriving scow will be
serviced by the second station; if both stations are occupied, the arriving scow is counted as
being in the queue.
4. The total service clock was reconfigured. Since it is possible that more than one scow is
unloaded at one station while the other station services a single scow, the length of time
added to the service clock is dictated by the station that is taking the longest.
5. The unloading time relationship was changed to the unloading time relationship shown on
Figure 5-6.
The best 30-day period of unloading that occurred during Phase 1 was from July 23rd through
August 22nd, when 69,300 CY of material were unloaded. The average removable load thickness
during this period was 2.0 feet, which is 0.3 feet greater than the average thickness of Phase 1.
The relationship between removable load thickness and unloading rate per foot was also
recalculated and is presented on Figure 5-6.
Figure 5-7 shows the result of this modeling exercise. The exercise included GE's specification
that a scow could not be loaded to a depth of more than 3.9 feet, since this would cause the scow
draft to exceed 8 feet which GE's dredging contractor asserts is the maximum stable draft.
Additionally, during Phase 1 it was a practice to place some amount of water over the sediment
in the scow to reduce air emissions, which also added to the draft of the scow and occupied some
of the load thickness that could otherwise have been occupied by sediment. Also as discussed
above, scows were not completely unloaded in Phase 1, so between 0.5 and 1 foot of material
remained in the bottom of the scow after unloading. This maximum loading depth (or thickness)
minus the amount of material left in the scow (0.5 feet) was used as an upper limit in the
modeling (3.4 feet maximum removable load); otherwise the simulated scow loads were drawn
randomly from the best 30-day of unloading during Phase 1 and increased by a specific thickness
up to a total thickness limit of 3.9 feet.
It is acknowledged that the true measure of productivity is the amount dredged from the river in
a given time period. During Phase 1, the total volume unloaded was about 12 percent less than
Hudson River PCBs Site Page 5-7
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
what was estimated to be dredged. The reason for this discrepancy is that some sediment is
removed as a slurry during dewatering prior to unloading scows and conversely some pore water
volume is counted as sediment during dredging and is included in the volume dredged.
However, the amount of material unloaded also reflects and is roughly proportional to the
amount dredged, when allowance is made for material removed as slurry when pumping water
from the scows. The volume number that is reported from the model is based on the simulated
scows arriving and unloaded (on a first-come, first-served basis) in a 30-day period. Since,
during Phase 1, scows were not unloaded on a "first-come, first-served" basis, the best 30 days
are off by a small amount and don't occur over the same exact period.4 Comparing the scow
arrival volume unloaded to the volume dredged, as noted, the amount unloaded is about 12
percent less, so if the scow arrival volume unloaded exceeds a productivity threshold, then it can
be assumed that the dredging volume also exceeds the target. The approximated dredging
volume is shown on Figure 5-7 as well.
Figure 5-7 shows the average results for both unloading with one station and with two stations as
a result of increasing the removable load thickness (750 runs per 0.25 foot increase to removable
load thickness represented by the data points). It also shows the one standard deviation greater
and less than the average for both the single station and the two stations simulations. Using a
single unloading station, the most optimistic monthly amount that could be unloaded is about
86,000 CY (one standard deviation above the peak average one-month productivity). This
equates to a dredging volume of just over 97,000 CY in 30 days. While this volume exceeds the
Phase 2 monthly target of 96,000 CY, it represents a level that cannot be sustained for the entire
Phase 2 program, based on the analysis here. The queue grows to an average of 12 scows when
1 foot is added (average of 2.9 feet removable load thickness). This shows that the unloading
station's capacity achieved during the peak 30 days of unloading in Phase 1 is not sufficient to
achieve the Phase 2 productivity target.
However, the Phase 2 productivity target can be met by adding a second unloading station and
increasing the scow loads by 0.5 feet above the Phase 1 average during the peak unloading
period to an average removable load thickness of 2.4 feet. This is an increase of 0.7 feet above
the average load thickness for all of Phase 1 and is well within the load window allowing for
material to be left in the scow after unloading and allowing water to be placed on the material
during transport. By adding a second station, the scow queue would be all but eliminated. The
benefits of reducing the queue will have ramifications at all points in the dredging program, such
as increasing the number of scows available for dredging and freeing up tugs that would
otherwise be occupied moving scows in the queue. This may also reduce vessel traffic and
reduce the time required to dredge a CU (and hence reduce the time an open CU is exposed to
the river flow), thus reducing resuspension.
The hand-written scow logs, included in GE's Data compilation Report as Appendix O - Barge
Trip Logs, have much more information in them than is presented in Appendix P. The
information in these logs could be used to further optimize the scow handling and unloading
system using the approach employed for this analysis. It is likely that, through such
4 This occurs because the volume in every scow load is different and if scows weren't unloaded in the order they
arrived then the unloading amounts and the arriving amounts are different and will peak at different times.
Hudson River PCBs Site Page 5-8
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
optimization, scow arrival rates could be increased, and that scow handling times could be
decreased. Together, these optimizations would result in significant increases in productivity.
References
Winston, W. L., Albright, S. C., 2009; Practical Management Science. Revised 3e; South-
western, Cengage Learning; ISBN-13: 978-0-324-66250-4; USA
Winston, W. L. 2004; Introduction to Probability Models. Fourth Edition; Brooks/Cole -
Thompson Learning; ISBN 0-534-40572-X; USA
Hudson River PCBs Site Page 5-9
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
5000
4500
4000
Scow queue peaks 7/24
begins decline 8/2
and does not go below 3 until 8/7
Daily scow loads ariving decline 7/31
Productivity peaks 7/30
and begins decline 7/31
Resuspension
events
¦ Scow Dailys Volume
o GE Daily Volume
Scows in Queue
a>
U
4 %
CO
£
txo
500
0 |*i i i i i i I i i i i i i I i i i i i i I i i i i i i I i i i i i i I i i ¦¦¦ i I i i i i i i I i i i i i i I i i i i i i I i i i i i i I i i i Piil| M i i i i | i i i A i | Pi i i i i | i Pi ill | i i*tt i i | i i i i i i | i i i i i i | i i i i i i | i i i i i i | ; i i i i i | i i ill i | i i i i i i | i i i
5/27 6/3 6/10 6/17 6/24 7/1 7/8 7/15 7/22 7/29 8/5 8/12 8/19 8/26 9/2 9/9 9/16 9/23 9/30 10/7 10/14 10/21 10/28 11/4
Note: Trend lines are running averages
Scow Queue at Loading Dock and Daily Dredging Productivity
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 5-1
April 2010
-------
60
50
40
30
a>
-Q
E
20
10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 3.1 3.2 3.3 3.4 3.5 3.7
Estimated Removable Load Thickness in Feet
Histogram of Scow Removable Load Thickness
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 5-2
April 2010
-------
Scow Removable Load Thickness (feet)
Relationship between Scow Load Thickness and Unloading Time
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 5-3
April 2010
-------
o
u
l/l
a>
-Q
E
80
70
60
50
40
30
20
10
a
. i. 1111.. 11.
i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i
0.1 0.3 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.5 4.5 5.5 6.5 8 9.5 11 14 16 19 27.5 40 51
Scow Handling Time (hours)
Histogram of Scow Handling Time
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 5-4
April 2010
-------
4.5
3.5
2.5
1.5
y = -0.0443x +0.3312
R2 = 0.0098
^ ~
~ ~
~
~ ~ ~
1— a ^ z—A
.*, V V^.S: n .«V~
0.0
0.5
1.0
1.5 2.0
Scow Load Thickness
2.5
3.0
3.5
4.0
Scow Removable Load Thickness and Arrival Rate
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 5-5
April 2010
-------
o
o
-------
150000
140000
130000
120000
110000
~ Two Stations
¦ One Station
• Phase 1 Average
A Phase 1 Best 30 Days Unloading
By loading 0.5 feet more than during Phase 1
and having two unloading stations, productivity is met
>-
u
¦a
0
CL
>
(0
T3
1
O
CO
At times it may be necessary to
load more to meet productivity goals _|_
Under optimal conditions, one station
does not meet productivity goal,
the length of queues limit productivity
"~T~-
1 standard deviation
50000
40000
30000
20000
Estimated dredging productivity
10000
1.00
1.50
+
+
2.00 2.50
Average Removable Load Thickness (feet)
3.00
3.50
Optimized Unloading System - Queuing Model Results
EPA Phase 1 Evaluation Report - Addendum - Hudson River PCBs Site
Figure 5-7
April 2010
-------
TOPIC 6
UNDERESTIMATION OF DEPTH OF
CONTAMINATION AND ITS IMPACTS ON THE
PROJECT
-------
TOPIC 6-A
POST-DREDGING CORE LOCATION
TREATMENTS AND THE NEED FOR MULTIPLE
RESIDUAL PASSES
-------
Topic 6-A - Post-Dredging Core Location Treatments and the Need for Multiple Residuals
Dredging Passes
Abstract
In the 10 CUs encompassing approximately 48 acres that were dredged during Phase 1, 445
locations were targeted for post-dredging sampling. Notably, less than 35 percent of the post-
dredging coring sites had a residuals layer requiring removal. By inference, less than 35 percent
of the Phase 1 dredging area was redredged to address a residuals layer. The purpose of this
relatively complex tracking and analysis performed by EPA was to measure the success of the
Standard in avoiding one or more residual passes, and to demonstrate that, contrary to what GE
has asserted in their Phase 1 Evaluation Report (GE, 2010), over 60 percent of the remediation
area went directly from inventory removal to compliant whereas only 8 percent of the Phase 1
area required 2 dredging passes. Moreover, since 6 inches of remaining inventory could not be
discerned from a true residual layer based on the Performance Standard sampling requirements,
the estimate of 35 percent is an upper bound on the actual extent of residuals. Ultimately this
analysis shows that the poorly defined DoC was the primary cause of additional dredging passes.
Specifically, 42 percent of the 445 locations were shown to have inventory present after the first
dredging pass based on identification of deeper contaminated sediments during one of the
subsequent sampling surveys and 20 percent of all locations actually required three inventory
passes.
Many of the post-dredging coring sites were occupied and cored several times. These core
collection efforts were the primary means to certify an area as compliant with the Residuals
Standard or to identify additional areas for residuals removal or inventory re-dredging. To
evaluate various reasons for re-dredging during Phase 1, the fate of each post-dredging core site
was tracked and post-dredging cores were grouped into six categories based on the treatment and
the depth of contamination (DoC). Clean cores {i.e., cores with TPCB less than lppm) and cores
that were considered compliant as per the residual standard were grouped together and
categorized as compliant.
The tracking and analysis of each post-dredging coring location yielded several important results
pertaining to the implementation of the Residuals Performance Standard:
• It identified areas where the DoC was poorly characterized and inventory remained below
the first dredging pass (volume removed by inventory dredging was more than double of
that removed by residuals dredging).
• It provided evidence that the standard was successfully implemented, without undue
difficulty or schedule impacts, and that dredging-generated residuals {i.e., residuals due
to spillage and bucket movements) were not a pervasive issue.
• It provided further support of EPA's proposal for an overcut addition to each dredging
pass in Phase 2 to address uncertainty.
6-A.l. Discussion
The Phase 1 areas encompassed approximately 48 acres, represented by 445 post-dredging
coring sites. Many of these sites were occupied and cored several times. These core collection
Hudson River PCBs Site Page 6-A-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
efforts were the primary means to certify an area as compliant with the Residuals Standard or to
identify additional areas for re-dredging. EPA's Phase 1 Evaluation Report summarized the
various dredging passes on the basis of core results and the associated dredge pass "footprint" as
a means to characterize the various types (e.g., inventory us. residual) and areal extent of re-
dredging required. However, this analysis did not track the treatment history of individual coring
sites (or nodes) and so it could not be determined how much of the river bottom required a single
residual pass (a removal of single layer of six inches of sediment) as opposed to areas where
multiple residual passes were necessary, presumably due to dredging-related spillage or
disturbance on the prior dredging pass. It was the intention of the Residuals Performance
Standard to avoid multiple dredging passes to remove residuals. This analysis is intended to
measure the success of the Standard in avoiding multiple residual passes.
To evaluate the various reasons for re-dredging during Phase 1, the fate of each post-dredging
core site was tracked through the Phase 1 program. Post-dredging cores were grouped into six
categories based on the treatment and the depth of contamination (DoC) in the post-dredging
core. Post-dredging cores with DoC greater than 6 inches were classified as additional inventory
and cores with DoC less than or equal to six inches were classified as residual. Core sites
requiring more than 6 inches of removal on any single pass after the first dredging pass when the
design cut lines have been attained are clear evidence of a poor local DoC and not sediment
residuals. By classifying cores in this manner, it is possible to obtain an upper bound on the
number of core sites where a residual layer was actually found. This means of classification
effectively includes any core locations where the true DoC was actually six inches beyond (i.e.,
deeper) than design cut lines. Even with this conservative assumption, the discussion below will
show that residual layer removal was not a major concern in Phase 1.Clean cores (i.e. cores with
TPCB less than lppm) and cores that were considered complaint as per the residual standard
were grouped together and categorized as compliant. As noted previously, in the 10 CUs that
were dredged during Phase 1, 445 locations were targeted for post-dredging sampling.
Tables 6-A-la to 6-A-lc list the status of each post-dredging sampling site after a dredging pass
and their subsequent grouping into one of six categories in the subsequent pass. Table 6-A-ld
provides the final status of the post-dredging coring locations. Note that these tables track the
status of the entire suite of coring sites after each dredging pass and not the cores themselves.1 In
each table, the left-most column summarizes the status of all 445 post-dredging sampling
locations after the prior dredge pass. The remaining columns track the fate of those locations that
required additional grouping into one of the six categories. Thus there are entries in the rows
labeled "Abandoned", "Inventory", "Residuals" but none in the rows labeled "Capped" or
"Compliant", since the latter categories do not require further dredging at the sampling locations
and hence are not subject to additional sampling.
For the purpose of illustration, a few of the location tallies will be traced through the first two
tables, beginning with the residuals. Summing across all CUs at the end of the first dredging pass
in Table 1-6-la, there were 115 sampling locations (out of 445) that were identified with a
residual layer of contamination (the italicized bold number in the first numeric column of the
1 Note that the second dredge pass in CU 1, which was only addressed a portion of the CU, was not included in the
tabulation. Thus only four dredging passes are presented in these tables.
Hudson River PCBs Site Page 6-A-2
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
table). Upon completion of the second dredging pass, these 115 locations were then re-sampled
and re-characterized as shown in the right side of the table. Of the 115 locations, 6 were reported
as not sampled (presumably due to lack of sediment), 4 were abandoned, 5 were capped, 64 were
compliant and required no further grouping into the categories, 7 were classified as additional
inventory (indicating there was significant uncertainty with the DoC at these locations and that
they were not truly residual locations as originally classified) and 29 were identified as still
exhibiting a residual layer.
Combining the 29 locations that still exhibited residuals with the other locations that exhibited a
residual layer after the second dredging pass2 yields a total of 62 locations with a 6 inch residual
layer after the second dredging pass. This is shown at the second column from the right in Table
6-A-la as well as in the first column of the Residuals row in Table 6-A-lb. The total from the
inventory column in Table 6-A-la transfers to the total in the Inventory row in Table 6-A-lb in
the same manner. For the Compliant tally in Table 6-A-lb, the total of 241 locations is the sum
of the Compliant column in Table 6-A-la (i.e., 123 new compliant locations) and the 118
compliant locations achieved after the first dredging pass (from the Total Number of Locations
column at the left of Table 6-A-la). The various tallies in each of the tables are constructed in a
similar fashion.
Table 6-A-la Treatment of Post-Dredging Sampling Locations after Second Dredging Pass
Post-Dredgii
Location Sta
First Dret
lg Sampling
tus after the
ging Pass
Post-Dredging Sampling Location Status after the Second Dredging
Pass - (320 locations were grouped into 6 categories)3
Classes
Total
Number of
Locations
Abandoned
Capped
Compliant
Inventory
Residual
Not
Sampled
Abandoned
33
9
2
2
1
3
16
Capped
7
0
0
0
0
0
0
Compliant
118
0
0
0
0
0
0
Inventory
169
2
11
57
62
30
7
Residuals
115
4
5
64
7
29
6
Not
Sampled
3
0
0
0
0
0
3
Total
445
15
18
123
70
62
32
2 Note that these additional locations were classified as inventory or abandoned at the end of the first dredging pass.
3 This total is calculated as the total number of sites less the sum of the number of locations previously capped and
the number of compliant locations.
Hudson River PCBs Site Page 6-A-3
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 6-A-lb Treatment of Post-Dredging Sampling Locations after Third Dredging Pass
Post-Dredging
Sampling Location
Status after the Second
Post-Dredging Sampling Location Status after the Third Dredging
Pass - (179 locations were grouped into 6 categories)
Dredging Pass
Total
Number of
Not
Classes
Locations
Abandoned
Capped
Compliant
Inventory
Residual
Sampled
Abandoned
15
0
0
0
2
3
10
Capped
25
0
0
0
0
0
0
Compliant
241
0
0
0
0
0
0
Inventory
70
4
11
20
25
10
0
Residuals
62
0
15
38
3
6
0
Not
Sampled
32
0
6
4
4
2
16
Total
445
4
32
62
34
21
26
Table 6-A-lc Treatment of Post-Dredging Sampling Locations after the Fourth Dredging Pass
Post-Dredging Sampling
Location Status after the
Third Dredging Pass
Post-Dredging Sampling Location Status after the Fourth Dredging
Pass- (85 locations were grouped into 6 categories)1
Total
Number of
Not
Classes
Locations
Abandoned
Capped
Compliant
Inventory
Residual
Sampled
Abandoned
4
1
3
0
0
0
0
Capped
57
0
0
0
0
0
0
Compliant
303
0
4
0
0
0
0
Inventory
34
1
33
0
0
0
0
Residuals
21
0
16
5
0
0
0
Not
Sampled
26
0
1
0
0
0
25
Total
445
2
57
5
0
0
25
Note:
1. The majority of the nodes requiring a fourth pass were associated CU-1 (41) which never had
a correctly defined DoC for the majority of the area and instead was completed by capping
the entire CU. This represents 41 of the 57 locations capped after the fourth pass.
Hudson River PCBs Site Page 6-A-4
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Table 6-A-ld Final Disposition of Post-Dredging Sampling Locations
Post-Dredging Sampling
Location Status after the
Fourth Dredging Pass
Classes
Total
Abandoned
27
Capped
114
Compliant
304
Inventory
0
Residuals
0
Total
445
These tables present the relatively complex tracking necessary to trace the fate of each post-
dredging sampling location. However, this analysis yields several important results pertaining to
the implementation of the Residuals Performance Standard. These tables show that of the 198
(115+62+21) instances where a residual layer was identified, only 35 (29+6) were re-dredged
again {i.e., a second residual pass) while seven residual cores indicated presence of inventory.
This also means that there were only 156 (198 - 35 - 7) locations4 that had a residual layer
identified. This means only 156/445 locations (35 percent) of the Phase 1 remediation area
yielded a residual layer. As noted previously, it is likely that a significant fraction of the 156
locations actually represent locations with underestimated DoC and not a true residual layer of
disturbed and spilled material from the original overlying sediment. The upper bound estimate of
156 locations out of 445 indicates that residual layer formation was not a significant problem for
the remediation. Of the 156 residual layer locations, only 35 required a second residual pass.
Another 23 (not shown directly in the tables) were capped after a single residual layer removal
attempt. This indicates that the majority of sites where a residual layer was identified (156 - 35 -
23 = 98 or 63 percent of these locations) were made compliant with just one more dredging pass
{i.e., a single residual layer removal pass).
In total, 65 percent of the post-dredging sampling locations were handled without the need of a
dredging pass targeting a residual layer {i.e., 65 percent of the area in Phase 1 did not require a
residual layer removal, a total of 289 locations). Either the area became compliant after
inventory removal (189 locations) or it was capped (78 locations). An additional 22 locations
were abandoned and not counted in these categories. It should be noted that in many cases core
locations required capping after incomplete dredging due to the impending end of the
navigational season; this was a direct result of underestimation of the depth of contamination.
These summations of the 6 category groupings of the post-dredging sampling locations show that
4 The 156 sites are the result of the following summation:
198 instances when a residual layer was identified
- 35 second residual pass sites (these are a subset of the sites where at least one residual layer was identified)
- 7 inventory sites (these sites were incorrectly identified as residual layer sites since the subsequent sampling
identified a foot or more of contaminated sediment)
156 sites where at least one residual layer was observed.
Hudson River PCBs Site Page 6-A-5
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
residual layer formation was not a significant issue for the Phase 1 effort. This is further borne
out by the large number of compliant locations (118 locations or 27 percent) achieved after the
first dredging pass. That is, 27 percent of the first inventory pass went directly to compliant
without requiring a residual layer removal. An even larger fraction (41 percent) of the later
inventory removal sites also went directly to compliant without requiring a residual layer
removal.
In contrast, the impact of the poorly defined DoC can be identified in these tables as well. After
the first pass, 169 of the 445 locations (38 percent) required inventory removal. Ultimately, 186
locations (42 percent) would be identified with inventory present.5 Of these 186 locations that
required 2 inventory passes, a large fraction (87 locations or 47 percent) actually required a third
inventory dredging pass. Thus 38 percent of the Phase 1 sites required 2 inventory removal
passes but 20 percent actually required three dredging passes. When the number of sites
requiring multiple inventory removal passes is coupled with the definition that each of these
locations had at least 1 foot of sediment removed on each inventory pass, it becomes clear that
the majority of additional dredging passes (and the majority of the additional dredging volume)
were needed to address remaining inventory and not a true residuals layer, again indicating that
the Phase 1 efforts did not expend a large effort "chasing" residual contamination.
Of the 35 locations where a second residual layer removal pass was required, only 13 required a
subsequent cap, indicating that the second residual pass was successful 63 percent of the time.
Taken together, only 35 of 445 post-dredging coring sites (8 percent) required 2 dredging passes
for residuals removal. In an additional 23 locations, (5 percent) capping was done after a single
dredging pass. Thus only 13 percent of the total number of post-dredging sites (and by inference,
about 13 percent of the area remediated) required any treatment beyond a single residual layer
dredging pass.
The very limited number of second residual layer removal passes indicates that the Residuals
Performance Standard did not require extensive residual dredging passes, and permitted capping
at locations where the CU was close to compliance. Based in this evidence, the standard was
successfully implemented, without undue difficulty or schedule impacts.6 Additionally, this
evidence shows that dredging-related residuals were not a pervasive issue. This analysis provides
further support for EPA's proposal for an overcut addition to each dredging pass. Since
dredging-related residuals were not problematic, an overcut included as part of each pass in
5 This value is obtained by summing across all the dredging passes as follows:
Sum of the sites identified with inventory on the second through fourth passes (169 +70+34) less the sum of the
inventory sites that were identified for inventory more than once (62+25) yields 186 sites with inventory identified
after the first dredging pass. Note that the later sum (62+25) represents the number of sites requiring three inventory
dredging passes.
6 The proposed changes to the Residuals Standard will tighten the requirements for capping to require at least one
dredging pass of 6 inches before capping to avoid the large number of locations that were capped without even a
single residual pass (only 30 percent of the locations capped had at least one residual pass prior to capping). While
this requirement might be viewed as increasing the number of passes, the addition of an overcut is expected to
reduce both the number of passes as well as the extent of capping by addressing the DoC uncertainty.
Hudson River PCBs Site Page 6-A-6
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
Phase 2 will help to address the DoC uncertainty, the major reason for re-dredging in Phase 1
and ultimately reduce the number and extent of additional dredging passes.
The Residuals Performance Standard helped identify areas where the DoC was poorly
characterized and where inventory remained below the first dredging pass. EPA's analysis shows
that poorly defined DoC was the primary cause of additional dredging passes, required to address
the remaining inventory and not to "chase" a residual contamination layer. After the first
inventory dredging pass, 38 percent of the 445 locations required subsequent removal of
inventory, while only 26 percent required a pass targeting residuals. Ultimately, 42 percent of
locations were shown to have inventory present after the first dredging pass and 20 percent
actually required three inventory passes. Taken together, the extensive evidence for inventory
(42 percent of post-dredging coring locations) plus the locations with a residual layer or simply a
thin remaining inventory layer (35 percent of post-dredging coring locations) indicate that an
overcut of at least 6 inches on the first dredging pass would have removed additional
contaminated sediment in 77 percent of locations, thus supporting EPA's recommendation for an
overcut addition to the design surface for Phase 2.
Hudson River PCBs Site Page 6-A-7 The Louis Berger Group, Inc.
EPA Phase 1 Evaluation Report - Addendum April 2010
-------
TOPIC 6-B
RELEVANCE AND CONSEQUENCES OF
UNCERTAINTY IN MEASUREMENTS OF THE
DEPTH OF CONTAMINATION
-------
Topic 6-B - Relevance and Consequences of Uncertainty in Measurements of the Depth of
Contamination
The relevance and consequences of uncertainty in the depth of contamination (DoC)
measurements used to design dredging cut lines have been discussed over the course of the
remedial design. This is exemplified by the comments and associated responses that were
exchanged between EPA and GE regarding the Intermediate Design Report (IDR) in December
of 2005. The specific document that captures this exchange is "Responses to USEPA Comments
on the Phase 1 Intermediate Design Report (IDR)", GE, December 26, 2005. Comments 37, 38
and 39 (pages 17 to 20) highlight the discussion of DoC uncertainty and are informative
regarding the respective stances adopted on the subject. In comment 37, EPA cautioned GE that
the uncertainty at individual core locations, which EPA estimated to be about 1 foot, would
outweigh GE's estimate that DoC measured in the cores would be conservative; EPA further
cautioned that "underestimating DoC may lead to additional re-dredging to remove inventory."
GE responded that the actual amount of inventory would be determined by cores collected during
dredging and that "the processing plant will have some capacity to dewater sediment generated
by those instances that DoC turns out to be greater than indicted [sic] by the SSAP data." GE
further emphasized their belief that the uncertainty was overestimated by EPA and stated that
high-confidence cores show a different story. However, as shown in Figure 6-B-l, high-
confidence core co-located pairs in Phase 1 have an average DoC discrepancy of more than
1 linches, confirming EPA's previous estimate.
In comment 38, EPA cautioned GE that conservative factors GE was maintaining cause the DoC
to be overestimated are not additive and cannot be used to show the DoC is overestimated. GE
reiterated their belief that these factors cause the DoC to be overestimated. In comment 39, EPA
stressed that GE's assertion that DoC is overestimated due to drag-down of contaminants during
coring is incorrect because it is a random event that will not affect a significant amount of cores.
GE maintained that it is a prevalent process that could have biased most cores, especially cores
with high peak concentrations. The results of Phase 1 dredging show that drag-down of
contamination during core collection did not affect the estimate of DoC.
In light of GE's concerns about cost-effectiveness, EPA accepted GE's positions and took an
adaptive management approach in approving the Phase 1 design. No overcut was required and
GE was given flexibility to manage the uncertainty in DoC through other means as they
implemented the project. However, uncertainty in DoC was not addressed during Phase 1
implementation, and its underestimation resulted in significant re-dredging. It is not appropriate
to consider the consequences of underestimating DoC as an unexpected impediment to
productivity, nor to assert that the need to re-dredge was caused by deficiencies in the Residuals
Standard.
Similarly to the exchange of comments and response presented on the 2005 IDR, advice was
given to GE in 2005 that the DoC was underestimated and would result in difficulties with the
residuals standard as well as re-dredging. In a memo attached as Appendix H to the Phase 1
final Dredge Area Design (DAD), GE was cautioned by EPA:
Hudson River PCBs Site Page 6-B-l
EPA Phase 1 Evaluation Report - Addendum
The Louis Berger Group, Inc.
April 2010
-------
"GE has reported that extrapolation of depth of contamination based on incomplete cores
is an important source of error in development of design elevations. EPA agrees that
DoC extrapolation is error prone and likely to be an important factor contributing to high
redredging rates, underestimation of the total in place volume and mass of contaminated
material and subsequent difficulties meeting residual standards."
The memo recommended further analysis of the extrapolation of incomplete cores through
comparison to co-located cores. This comparison and refinement of extrapolation was not done
prior to design of the Phase 1 cut lines. As a result, consistent underestimation of the DoC in
developing design cut lines resulted in significant re-dredging to address previously unidentified
PCB inventory.
In keeping with the adaptive management approach, it is clear that a re-evaluation of the process
for determining the DoC and establishing design cut lines, including an overcut to address
uncertainty, is required. It is expected that a process of continual evaluation and adjustment will
be continued throughout the dredging project.
Hudson River PCBs Site Page 6-B-2 The Louis Berger Group, Inc.
EPA Phase 1 Evaluation Report - Addendum April 2010
-------
-------
TOPIC 2 ATTACHMENT
CONDITIONS ASSOCIATED WITH WATER
COLUMN PCB CONCENTRATIONS:
THOMPSON ISLAND DAM 2009
-------
Topic 2 Attachment
CONDITIONS ASSOCIATED WITH
WATER COLUMN PCB
CONCENTRATIONS:
THOMPSON ISLAND DAM 2009
Multivariable Analysis of
Water Column PCB and Operational Data
April 23, 2010
Prepared by:
KERN Statistical Services, Inc.
Prepared for:
United States Environmental Protection Agency
Region 2
290 Broadway
New York, NY 10007-1866
Under Contract to:
Louis Berger Group, Inc.
412 Mt. Kemble Avenue,
Morristown, NJ 07960
KERN Statistical Services, Inc.
5175 NE River RD
Sauk Rapids, MN 56379
-------
1.0 SUMMARY
It has been stated by GE that water column concentrations can be "Predicted based only on PCB
removal rate and river velocity." Based on this conclusion and on application of such a model to
the Phase 1 results, GE broadly concluded the following in the Resuspension portion of their
February 15, 2010 presentation to peer review panel:
1. Phase 1 demonstrated that the Resuspension Standard cannot be met (ES-5), and that
EPA's proposed increase to the load standard would compromise the remedy (ES-7).
2. The federal drinking water standard of 500 ng/L PCBs will be exceeded frequently in
Phase 2 because the rate of dredging will exceed that of Phase 1, and fewer
resuspension control opportunities will be available.
3. Re-deposition in non-dredge areas will compromise remedy benefits.
4. There are no practical means to reduce resuspension and avoid exceeding the
Resuspension Standard.
2.0 PRELIMINARY FINDINGS
Subsequent to releasing these results, a preliminary analysis of factors associated with water
column PCB concentrations was conducted and described in this report. It was found that while
water column PCB concentrations are indeed positively associated with mass of PCBs removed,
as indicated by a Spearman Rank Correlation coefficient of 0.42, a more detailed analysis
suggests that this relationship is due to a combination of several operational factors, some of
which are readily manageable in ways that would logically be expected to reduce PCB releases
associated with dredging operations. Spearman Rank correlation coefficients are shown in Table
1.
Based on the analysis reported here, it can be concluded that the mechanisms associated with
increased water column PCB concentrations are numerous, varied, and should not be simplified
to a mere proportionality of the mass removed, as suggested by GE. Mass and volume removed,
which are both functions of the direct measurement, bucket counts, is a surrogate, integrating the
net effect of all of the processes involved in dredging, and therefore correlates well with water
column PCB concentrations. However, this is not sufficient to conclude that individual
operational variables cannot be managed to reduce resuspension of PCBs during Phase 2
dredging. The manner in which dredging and other related activities were conducted during
Phase 1 is very important and provides a good source of information on how dredging and other
activities could be modified to reduce resuspension of PCBs during Phase 2.
Based on the multivariate analyses of daily processes and water column data, water column PCB
concentrations are found to be positively associated with several factors, all of which are
expected to influence release and resuspension of PCB contamination. These factors include:
KERN Statistical Services, Inc.
1
-------
1. Sediment removal {i.e., bucket counts, volume removed, mass removed).
2. Flow rate.
3. Vessel traffic (primarily the distance traveled by scows).
4. The number of CUs being backfilled in any given day.
5. The area and concentration of freshly disturbed sediments in CUs open to the water
column each day.
6. Bucket cycle time and volume removed per cycle termed the apparent fill-rate (fill-
rate) and other surrogates to sediment spillage.
Thirteen of 28 process variables (shown in Tables 1 through 3) considered in this analysis
demonstrated statistically significant positive associations with water column PCB
concentrations, with squared Spearman Rank correlation coefficients ranging from
approximately 0.2 to 0.4. These levels of association are individually weak, indicating that no
single process can be identified as the source of resuspension, but rather that a complex set of
interactions among processes is required to describe water column PCB concentrations. The well
known adage is that "association does not imply causation" but, conversely, lack of apparent
association also does not eliminate causation. It is expected that multiple variable models may be
necessary to adequately explain variation in water column concentrations. It is shown that there
were periods of time in Phase 1 when resuspension rates were low (less than 200 ng/1) and
productivity was adequate to meet proposed Phase 2 productivity criteria.
It is further expected that controlling PCB resuspension may be accomplished through a
combination of additional or improved management practices and strategies applied during
several stages of the sediment removal and backfilling process. The following sections describe
the modeling approach used in this analysis and the resulting relationships between operational
variables and water column concentrations at TID.
3.0 OBJECTIVE
The primary objective of this analysis is to develop an empirical, data-driven model describing
water column concentrations of Total PCB as a function of the physical and operational variables
associated with remedial activities conducted during Phase 1.
4.0 INTRODUCTION
Sediment transport investigations are often characterized by limited data, necessitating the
development of theoretical equations {e.g., models) to describe the fate and transport of
contamination between the sediment and the water column.
In contrast, during Phase 1, dredging, data quantifying salient features of the sediment removal
process were recorded on daily, weekly and at times fifteen-minute sampling intervals—e.g.,
KERN Statistical Services, Inc.
2
-------
vessel positions were recorded every fifteen minutes throughout the Phase 1 project. These data
are certification-unit specific, available on a daily basis, and can be tied to daily water column
concentrations monitored at near- and far-field stations in Thompson Island Pool. They provide
the basis to develop an empirical model of water column concentration as a function of measured
data specific to the operations in the Thompson Island Pool during Phase 1. A combination of
Factor Analysis (Seber, 1977) and Multiple Regression (Neter el al., 1996) was used to
statistically identify groups of parameters most strongly associated with water column PCB
concentrations. This empirical approach provides the opportunity to test hypotheses and
assumptions that otherwise would remain untested in more typical situations where data are
sparse. These data provided the opportunity to evaluate the influence of various remediation
processes on water column PCB concentration.
5.0 METHODS
5.1 Data
Operational data were collected throughout the Phase 1 season which quantified the primary
aspects of dredging operations. In addition, water column Total PCB concentrations were
measured daily, providing the potential to develop a retrospective model describing the
relationship(s) between mechanisms of the dredging process and water column Total PCB
concentrations. The data obtained includes metrics quantifying potential sources of PCBs
associated with:
1. Flow and temperature conditions;
2. Debris removal;
3. Volume and mass removed;
4. Sediment disturbance associated with prop wash from vessel traffic;
5. Efficiency of removal operations—incomplete closure of buckets, cycle times;
6. Resuspension of exposed PCB deposits in open CUs; and
7. Sediment disturbance associated with backfilling.
In all, 28 variables were statistically tested for the potential to predict water column PCB
concentrations at far-field stations located downstream of dredging operations. Data were
summarized on a daily basis, so for most variables there were approximately 166 days (May 15
through October 27, 2009) for which PCB concentration could be compared with dredging,
debris removal and backfilling process variables.
Some variables quantifying how dredging was conducted, such as bucket closure and apparent
fill rates, were only available for the 127 days within that 166-day time period when active
dredging occurred. Therefore, the analyses were conducted for variables measured on all 166
KERN Statistical Services, Inc.
3
-------
days and then repeated for the subset of variables measured on the 127 days when active
dredging was occurring, primarily after June 2009. Data collected during active dredging (i.e.,
127 days) are more closely associated with how dredging was conducted and are therefore the
primary source of information on how dredging and other activities could be modified to reduce
resuspension of PCBs to the water column.
5.2 Modeling Overview
The factor analysis and regression approach shown in Eq. (1), below, was used to develop a
model of the form
Cyvater ~ Cbaseline ^l^Source-1 K2^Source-2 Kn^Source-n (1)
where:
1) the constants Kh K2, K3,... K„ are loosely interpreted as "nef sediment-to-water
partitioning coefficients for each source, and
Cyvater ~ Cbaseline ^l^Source-1 K2^Source-2 Kn^Source-n (1)
Phase 1 is unique in the extent of data available to not only estimate these coefficients, but also
to identify combinations of measured processes that are most important for predicting water
column concentrations. Multiple regression analysis was used to identify metrics contributing
significantly to prediction of water column PCB concentrations. The fitted model can also be
used to identify combinations of process variables characteristic of periods of time when
resuspension was low and productivity was high.
Surrogates and Confounding
Metrics described here should be considered surrogates for the physical processes of interest.
For example, it is not clear that there is a partitioning coefficient between bucket counts and
water column concentrations, however if a mechanistic model relating sediment losses to bucket
count were developed, there would be a "net" partitioning of PCBs relating water column
concentrations to sediment losses based on bucket counts. It is this net partitioning that is
estimated by the coefficients of the regression model.
Similarly, because the mass removed per day is derived from bucket counts and other variables,
it would also be expected that water column concentration would be correlated with mass of
material removed per day. In fact, mass removal is clearly a surrogate integrating most
processes likely to cause PCB losses and resuspension. Unfortunately this does not provide
useful insight into how operational processes can be modified to reduce PCB sources to the
water column. Any variables that are correlated to material disturbance and removal would be
expected to correlate with water column concentration. Additionally, some of those variables are
also expected to be inter-correlated amongst one another. In statistical terms these variables
would be considered "multi-collinear."
KERN Statistical Services, Inc.
4
-------
Cause and Effect
Because the data developed through this study are observational rather than based on a designed
experiment, and because many of the process variables are inter-correlated, cause and effect
relationships cannot be inferred directly. It is important to consider variable relationships to be
associative rather than causative unless other lines of evidence can be used to eliminate other
plausible causative processes. This is due to the surrogate nature of most of the metrics. For
example, correlation between bucket counts and water column concentrations could lead one to
conclude that use of larger buckets would reduce the daily bucket count per unit volume
removed and, by extension, water column concentrations. However, if the actual acting
mechanism is due to disturbances from scow traffic, which also would be expected to be
associated with bucket counts (e.g., more buckets produce more volume requiring more scow
trips), then modification to the bucket size in efforts to reduce bucket counts would be futile.
The data are observational and therefore their use in developing best management practices
should take into account that individual metrics may be surrogates for what may be lurking un-
qualified processes.
Bivariate Correlation Analysis
The first step in the regression analysis was to analyze the pairing of each individual process
variable with water column PCB concentration to test for a positive association. A bivariate
relationship between a process variable and water column PCB concentration is indicative of at
least a surrogate relationship that could potentially be causative and should be considered for
plausibility.
These bivariate relationships were summarized by calculating Spearman Rank correlation
coefficients between water column PCB concentrations and each process variable of interest.
Correlations were also calculated for water column concentrations lagged by 1 and 2 days to
determine if subsequent analyses should incorporate adjustments for travel time between the
dredging areas and the far-field monitoring stations.
5.3 Multiple Regression
A fundamental assumption of multiple regression is that the predictor variables (i.e., source
terms) are statistically independent. In this situation, it is clear that many of the source terms of
interest are inter-correlated, e.g., more sediment volume removed requires more vessel traffic.
Therefore some groups of source terms cannot be entered directly into multiple regression
models without careful consideration of their interrelations. As indicted above, these related
predictor variables often termed "multi-collinear." A great deal of effort has been devoted to the
study of the effects of multi-collinearity and methods to mitigate these effects on the
interpretability of model coefficients and the predicted values.
KERN Statistical Services, Inc.
5
-------
Regression models are typically used for two purposes: 1) prediction of the response variable,
and 2) testing and interpretation of the regression coefficients. For prediction, one would be
primarily interested in estimating future water column concentrations under a set of conditions
previously measured in the model fitting process. As long as the future conditions are within the
range of the variables used to estimate the model coefficients, multi-collinearity generally does
not adversely impact predictions. However, multi-collinearity does make interpretation of
causative relationships, such as the importance of dredging-related vessel traffic, more difficult.
Careful model construction and evaluation of sub-models to deduce the plausibility of causative
processes is necessary.
Recognizing the limitation of multiple regression models in the differentiation of individual co-
linear factors, this analysis is designed to develop a predictive model and to qualitatively
evaluate the relative importance of the factors associated with water column total PCB
concentrations. From this analysis it is possible to identify groups of process variables
collectively associated with changes in water column PCB concentrations. Identification of such
independent process variables suggests which components of the dredging process should be
investigated for potential causative relationships.
5.4 Factor Analysis
In factor analysis, a set of statistically independent predictor variables are derived from the
collection of 28 inter-correlated variables. These independent variables are obtained by applying
a factor analysis to the full collection of predictor variables and deriving surrogates for the
dredging processes that are statistically independent and can be entered jointly into multiple
regression models. Factor analysis is similar to principal components analysis, with the
exception that the principal components are "rotated" through an orthogonal transformation that
often results in component loadings (also called factor weights) that are more physically
interpretable.
It is recognized that there are no unique or optimal factor solutions. However, development of
independent scores that are composed of interpretable groups of process variables is desirable for
the purposes of developing a predictive model, as well as for interpretation of the relative
importance of independent groups of process variables. It is fully recognized that because many
process variables are inter-correlated, fully dissecting the relative importance of each process
variable may not be possible. However, to the extent that factor scores can identify independent
groups of variables, the contribution of each group can be distinguished through this approach.
Groups of variables providing redundant information are grouped together and these groups of
variables form independent composite variables called factor scores. These factor scores are
statistically independent and often are physically meaningful.
KERN Statistical Services, Inc.
6
-------
Regression on Factor Scores
The results of the factor analysis were used to transform each group of daily process variables
into a linear combination of independent variables called factor scores. These factor scores have
the advantage of being statistically independent and are therefore compatible with the
assumptions of multiple regression. A multiple regression was used to identify those factors that
were important to prediction of water column PCB concentration. Important factors were
defined as those factors with regression coefficients that were significantly different from zero at
the 5 percent level of statistical significance. The resulting model is suitable for use as a
predictive model and by inspection of the factor loadings, these models can also be used to
identify independent combinations of process variables important to prediction of water column
PCB concentrations. Effects due to variables nested within a common factor are difficult to
distinguish without other lines of evidence.
The results of this analysis can also be used as a guide in the development of mechanistic models
that are specific to individual process variables. In particular, when more process-oriented
variables are to be calibrated against water column data, their settings should be consistent with
the interrelations found here through factor analysis. Unless certain processes can be eliminated
through other data and analysis, it would not be reasonable to assume that individual process
variables can be eliminated purely through identification of other surrogates that are more
strongly associated with water column concentrations.
For example, mass removed per day can be tracked relatively accurately, while losses from
bucket lifts are much more difficult to measure directly. Therefore, the quality of mass removal
data is expected to be much less variable and therefore more likely to correlate with water
column concentration than sediment losses. Because of this difference in measurement quality
among variables, mass removal per day could appear to be the better predictor of water column
concentrations; however, this does not eliminate the potential that loss percentages might be the
more important process contributing to water column PCBs. The well known adage is that
association does not imply causation, but conversely, lack of apparent association also does not
eliminate causation.
Distinguishing independent root causes of water column concentration would require extensive
and careful multiple variable analyses combined with professional judgment and development of
mechanistic models in order to develop sound best management practices. The analysis
presented in this section is a first step in this direction, intended to provide an indication of the
major groups of processes influencing water column PCB concentrations. Until more detailed
sub-analyses are conducted it would be premature to eliminate any process variables from
consideration for improvement and refinement.
Regression on Individual Mechanisms
KERN Statistical Services, Inc.
7
-------
Information provided by GE (2010) correctly states that water column PCB concentrations at
TID increase in apparent response to the mass of PCBs removed from the Thompson Island Pool.
In order to improve the dredging process, it is necessary to identify the specific components of
the dredging operation that are contributing to increasing water column concentrations
downstream of dredging. An adequate understanding of the contributors to resuspension must be
developed in order to identify processes that can be adjusted so the resuspension standard can be
met with reasonable productivity levels during Phase 2 dredging.
Analysis of the many activities that occurred during the dredging season suggests that multiple
factors are likely to have contributed to increases in PCB concentrations at the TID and, if
identified, would result in a management opportunity to reduce the resuspension of PCBs.
Results of the multivariate factor analysis were used to identify process variables that would be
likely candidates for development of a predictive model of water column PCB concentrations
that would be an interpretable function of a minimal number of process variables. This more
refined model would provide a valid equation for prediction of water column PCB
concentrations at the TID under a range of process variable settings.
Because of the interrelationships among predictor variables, reliability of predictions can only be
assured for combinations of process variables that were actually used in the model fitting. For
example, one could not simply make forecasts based on reduced tug-miles without incorporating
commensurate reduction in scow availability and ultimately limiting productivity. Conversely, if
reduced tug miles were mitigated by filling scows more fully, to maintain productivity rates, this
could be considered a plausible approach. The regression model is most reliable for
combinations of predictors within the operational range of values observed in the data used to fit
the model.
The multivariate factor analysis was based on groups of individual variables summarized for the
entire Thompson Island Pool on a daily basis. For example, the distance traveled by particular
vessel types was represented as total distance per day without regard for the particular CU or the
concentration or depth of water over which the vessels were operated. Upon seeing the potential
for these simple surrogates to predict water column PCBs, additional spatially-explicit metrics
were developed and included in the multiple regression analyses.
Many of the mechanisms that are associated with water column PCB concentration that have
been identified through factor analysis are inter-correlated. Development of a meaningful model
based on a small number of interpretable mechanisms requires careful selection of combinations
of variables such that interrelationships do not degrade parameter estimates. As a more
interpretable alternative to factor scores, this can also be achieved by introduction of nonlinear
aggregate functions of the individual variables, and by careful selection of combinations of linear
and non-linear functions. Degradation of model reliability can be controlled by ensuring that
selected combinations of variables do not cause large variance inflation factors (VIFs). VIFs are
KERN Statistical Services, Inc.
8
-------
1.0 when predictors are statistically independent and increase with increasing levels of inter-
correlation of predictors. In general, VIFs less than approximately 5 are considered acceptable
and less than 2 are preferred. In this analysis, variable combinations were rejected when VIFs
exceeded approximately 2.
For each vessel type and within each certification unit (CU-0 representing non-CU areas), the
distance traveled was multiplied times the bed sediment concentration and divided by the
approximate water depth, defined as the 117.5 elevation minus the bathymetric survey elevation.
This variable is large for vessels that traversed shallow areas with high PCB concentrations, and
was low for vessels that were either in deep water, or traversed low bed sediment concentrations.
These metrics were also modified by setting values to zero whenever the water depth was greater
than 11 feet, signifying an approximate water depth beyond which resuspension due to propeller
wash is not expected to contribute on a consistent basis. For each vessel type, these data were
totaled for each CU on each day of the construction season.
Based on discussions between EPA and its contractors, another variable was developed based on
the number of scows waiting to be unloaded. It was observed that when scows were scarce,
dredging contractors tended to nibble at the sediment, increasing the number of bucket bites per
yard of sediment in apparent efforts not to fill scows prior to arrival of a replacement. It is
thought that dredging contracts may have incentivized this behavior. This nibbling tends to
cause greater disturbance and increased PCB resuspension. So, the number of scows on queue
was included in this analysis.
Variables added to this analysis that were deduced from the factor analysis, professional
judgement or field observations, include:
1. Average number of scows on queue per day. When scows are queued, dredging
contractors slow down and potentially nibble sediment, increasing the resuspension losses
per unit sediment removed.
2. (Volume removed)/(bucket count x bucket volume) x (mass removed). Index to spillage
associated with resuspension due to failure to fully close buckets in efforts to increase
removal rates. Failure to close buckets was signaled in the database by apparently over-full
buckets due to volume associated with multiple actual cycles being associated with a single
apparent cycle in the data.
3. Flow at Fort Edward. This is not a source of resuspension in and of itself, however
activities associated with dredging may have decreased critical shear stresses of the worked
sediment, resulting in generally increased resuspension of PCBs with increased flows.
4. CU and non-CU specific concentration and water depth-weighted tug distance. Previous
analyses looked at total vessel distance per day, but these totals were correlated with total
KERN Statistical Services, Inc.
9
-------
mass and volume removed. Separate CU-specific and concentration- and depth-specific
information was needed in order to disentangle vessel traffic from apparent effects of mass
and volume removal
5. (Product of number of units being backfilled) x (metric described in 4 - tug
disturbance).
6. CU-specific mass of PCBs removed. By separating mass removed per CU specifically, it
was possible to separate general removal effects from other factors
GE has hypothesized that PCB resuspension is purely a function of mass removed and water
velocity. To test this hypothesis, a variable was developed from the product of the mass
removed and the average apparent bucket fill rate. It was expected that greater amounts of
spillage would begin to occur when this apparent fill rate increased above approximately 60
percent. To test these competing variables, regression models were developed separately with
and without each variable and model fit was compared to evaluate the strength of evidence for
each hypothesis.
Bucket count data were used to calculate total mass removed from each CU per day. These
summaries were entered into regression models separately for each CU and in combinations
across groups of CUs in an effort to test the hypothesis that losses are proportional to mass
removed. This particular part of the analysis focused on groups of CUs in the East and West
Rogers Island areas and CUs 17 and 18 near East Griffin Island. Because CU-18 was divided
into an area enclosed by sheet pile (CU-18IN) and another area outside the sheet pile enclosure
(CU-180UT) the relative influence of mass removal could be compared with that due to tug and
scow traffic servicing these CUs.
6.0 RESULTS
Thirteen variables representing five groups of processes were identified to be associated with
water column PCB concentrations at the TID. Five variable groups (factors) were identified that
collectively explained 55 and 60 percent of the variation in water column PCB concentration in
the 166-day and 127-day models, respectively. These factors represented volume and mass
removed and efficiency, area of recently disturbed sediments in open certification units, vessel
traffic and backfilling. Following is a summary of the results of the analysis.
6.1 Bivariate Correlations
Squared Spearman Rank correlation coefficients for water column PCB concentration with each
of the process variables are reported in Table 1. The analysis was repeated with process
variables paired with one-day and two-day lagged PCB concentrations to evaluate the potential
impact of travel time on the strength of the correlation. These squared correlation coefficients
represent the proportion of variation explained by the relationship between water column PCB
KERN Statistical Services, Inc.
10
-------
concentration and each variable, analogous to an R2 from a regression. The Spearman
correlation coefficient is preferred because the assumption of linearity inherent in the Pearson's
coefficient is relaxed. The results summarized in Table 1 show that:
1. Correlations between water column PCB concentrations and process variables are
generally weak, ranging from 2 percent for debris removal to 42 percent for volume
and mass removal, indicating that no single variable could be expected to adequately
explain the fluctuations in water column PCBs observed during Phase 1.
2. Correlations for water column concentrations lagged by one day were less than those
for concurrent measurements, and two-day lagged measurements produced still lower
correlations.
a. In contrast, GE asserted that weekly averages were needed to counter the
effects of travel time in their analysis of the water column PCB data.
b. This weekly averaging approach is counterproductive given the lack of
correlation between lagged water data and process variables.
c. Weekly averaging would artificially reduce the power to detect subtle
multiple-variable relationships, suppressing potentially important relationships
between water column PCB concentration and operational variables.
3. Statistically significant positive associations were identified for most processes
expected to disturb sediments.
a. Volume and mass removed (R2=0.22 to 0.42).
b. Dredging efficiency measures such as apparent bucket fill rate and depth of
cut (R2=0.08 to 0.22).
c. Sources due to area of open CUs (R2=0.15 to 0.19).
d. Debris removal (R2=0.02).
e. Boat traffic (R2=0.11 to 0.26).
f. Backfilling operations (Number of CUs being backfilled) (R2=0.09).
4. Weak statistical relationships may be indicative of surrogate relationships that are
markers for important, but crudely quantified, sources of PCB resuspension.
5. Water column concentrations were negatively associated with flow at the Fort
Edward Station, but the relationship was not statistically significant.
6.2 Regression on Factor Scores - Factor Analysis
Because water column PCB concentrations were weakly associated with several operational
variables, efforts were made to develop a multivariable model that would adequately explain
water column PCB concentrations. Because several process variables were derived from basic
KERN Statistical Services, Inc.
11
-------
measurements such as bucket counts, it was expected that many process variables would be
inter-correlated. In order to understand interrelationships between process variables, a factor
analysis was conducted to identify a set of independent factors that would be both meaningfully
interpretable, as well as provide inputs for a regression model predictive of water column PCB
concentrations.
The factor analysis was conducted with only the predictor variables identified using data
obtained on all 166 days and that retrieved on just 127 of the 166 days. Resulting factor scores
were used as predictors in a regression analysis to identify important factors for prediction of
water column PCB concentrations.
Factor Analysis (127 day model)
Table 2 and Figure 1 show the factor weightings (also called loadings) for each of the 28 process
variables under consideration. The factor weightings are unitless, range from plus one to minus
one, and are considered meaningful when they exceed approximately 0.4 in magnitude.
Loadings that are less than 0.4 in magnitude are within the opaque rectangular area in Figure 1.
Cells in Table 2 are shaded green to draw attention to loadings that exceed this nominal level.
There were five factors associated with water column Total PCB concentration which described
from 2 to 37 percent of the total 60 percent variance in water column PCB concentration
explained by the regression model. Regression coefficients, standard errors, variance inflation
factors, and partial R2 values are presented in Table 4.
Factor-1 includes weighting on bucket counts, mass removed, volume removed, residual Total
PCB concentration in open CUs, and the product of mass and removal efficiency (ME). This
factor summarizes potential PCB sources from variables that are directly related to sediment
removal, as well as efficiency of the removal process.
Factor-6 is most heavily weighted on the amount of backfilling being conducted and the product
of flow and backfill (a surrogate for load from backfilling). This factor also has substantial
negative loadings on bucket counts and temperature. This may reflect that bucket counts
coincidentally varied inversely with temperature and backfilling operations, {i.e., It is
coincidental that dredging started slowly when temperatures were colder, increased through the
summer and then declined in the fall when temperatures also declined and backfilling
commenced).
Factor-7 is most heavily weighted on the area of open CUs and the flow- and concentration-
weighted surface area of open CUs. This factor has a clear signal exclusively related to the
amount of open CUs at any point in time that is independent of volume and mass removal.
Factor-8 is most heavily weighted on flow, and the product of flow and total vessel traffic. This
factor is also independent of variables in Factor-1 (describing removal metrics), indicating that
KERN Statistical Services, Inc.
12
-------
there may be an independent PCB source to the water column associated with vessel traffic.
This variable is a crude measure of potential sources due to vessel traffic, as it does not account
for either water depth or the concentration of areas over which traffic occurs. It is expected that
refinement of this variable will substantively improve its relative strength as a predictor of water
column concentrations.
Factor-9 is most heavily weighted on boat distance, which is a single metric that only accounts
for distance traveled by vessels.
As a general observation, these results show that resuspension of PCBs to the water column is
associated with a combination of removal activities, backfilling activities, vessel traffic, and the
surface area and duration that disturbed residuals are exposed in open CUs. This indicates that
best management practices could be applied to one or several of these processes to reduce
concentrations of PCBs in the water column.
Factor Analysis (166 Day Model)
The factor loadings for the 166 day model are summarized in Table 3 and Figure 2. Cells in
Table 3 are shaded green to draw attention to loadings that exceed the 0.4 nominal level for the
magnitude of factor weights. Factor groupings were similar because the majority of data were
common to both models. The model fit was slightly weaker, with an adjusted R2=0.55 as
compared to the R2=0.60 for the 127 day model.
The five factors associated with water column Total PCB concentration were qualitatively
similar to those identified in the 127 day model, representing variables associated with sediment
removal (semi-partial R2=0.28), backfilling (semi-partial R2=0.06), concentration weighted
surface area of open CUs (semi-partial R2=0.04), flow times vessel distance (semi-partial
R2=0.05), and mass removed (semi-partial R2=0.13). Because performance data relating to
apparent bucket filling rates were not included in the 166 day model, the separation of variables
among factors was less obvious and general surrogates for overall activity such as mass and
volume removal and boat traffic tended to group together in the first factor. This lower quality
model fit when dredging performance variables are not included shows that further investigation
of the processes controlling fluxes of PCBs to the water column should focus on variables that
characterize how dredging and other supporting operations are conducted rather than just on how
much sediment is dredged.
Fitted Factor Based Regression Model
The fitted regression model results are plotted in Figure 3 where the 127-day model is plotted for
the days on which all necessary variables were measured, and on the remaining 29 days the 166-
day regression modeled values are substituted. The fitted values show that the predictions track
the primary day-to-day fluctuations in concentration when dredging activities are in progress,
and in October, when GE's simpler model did not predict disturbance well. It is believed that
KERN Statistical Services, Inc.
13
-------
incorporation of performance data as well as variables quantifying backfilling and areas of CUs
contributes to the improved model fit. Estimated regression coefficients, standard errors, partial
R2 values, and variance inflation factors are summarized in Tables 4 and 5. The improved model
fit shows that GE's assertion that concentrations are driven exclusively by the amount of
dredging may not be fully justified and that improvements of component process associated with
dredging and backfilling would be expected to reduce resuspension of PCBs to the water
column. Also Upper 95 percent prediction limits are shown on Figure 4. These limits represent
an added benefit of the regression approach to model development in that uncertainties in model
predictions are made explicit.
6.3 Regression on Mechanistic Variables
The model fit was improved over the factor based models through directly regressing on selected
mechanistic variables that were summarized specific to individual CUs. The adjusted R2 for the
127 day model increased from 0.60 to 0.68with the primary improvements in model fit due to:
1. Incorporation of the dredge efficiency term in place of the more general mass removed
term proposed by GE;
2. Differentiation of vessel traffic by water depth and sediment bed concentration; and
3. Incorporation of a source term specific to activities inside the sheet pile wall at CU-18.
Primary Mechanisms Influencing Water Column PCB Concentrations
The multiple regression model for water column PCB concentration was developed using six
individual measured variables rather than factors, as described above. The overall proportion of
variation explained by the model was 0.68, with variable-specific proportions as follows:
1. Mass removal efficiency (partial R2=0.25).
2. Boat traffic specific to backfilling operations (partial R2=0.11).
3. Boat traffic specific to mass removal at CU-17 and CU-18 (partial R2=0.01).
4. Flow at the Fort Edward station (partial R2=0.08).
5. The number of scows on queue for unloading (partial R2=0.01).
6. Mass removed from inside the sheet pile enclosure at CU18 (partial R2=0.03).
When predictors are inter-correlated, the sum of the variance explained by each variable
independently is often less than the total variance explained {i.e., adjusted R2). This component
of the total variance is the variance explained jointly by the collection of variables in the model.
In this case, 19 percent of the variance was explained jointly, indicating that approximately a
KERN Statistical Services, Inc.
14
-------
quarter of the total variance explained by the model could not be uniquely tied to any single
measured factor. These results are summarized graphically in Figure 5.
Figure 6 shows the fitted model plotted against time, along with the observed water column
concentrations and upper 95 percent confidence limits for predicted PCB concentrations. Figure
7 shows the observed values plotted against the predicted values. This model, based on
mechanisms, provides better agreement between observed and predicted values, particularly
during early August and early September when high PCB concentration were observed at TID.
The mass removal efficiency variable was the primary predictor of water column concentrations
explaining 24 percent of the total variance in water column PCB concentrations. This variable is
comprised of the product of mass removed and apparent bucket fill rate as described above. This
variable identifies situations where mass removal is conducted when daily bucket fill rates are
calculated as greater than 60 percent and at times 150 percent (thought to represent situations
where multiple bucket bites are counted as a single bite due to debris or less-well controlled
operations).
This correlation between dredging efficiencies and water column concentration was evaluated to
determine if the correlation could be due primarily to correlation with mass removed. This
hypothesis was tested by comparing model fit for a model including mass removed alone with a
model including mass removal efficiency. The model including the other four variables
identified above plus mass removal efficiency had an adjusted R2=0.68, whereas the model
including the same four variables and just mass removed (i.e. without multiplying by efficiency)
had an adjusted R2=0.60. This indicates that mass removal efficiency was a better predictor of
water column concentration than mass removed alone. This relationship indicates that, in
addition to simple proportionality to mass removed, unclosed dredge buckets and potentially
other inefficiencies also contributed to resuspension of PCBs to the water column.
Disturbance due to tug traffic in shallow areas between CUs 17 and 18 and the Lock 7 entrance
was found to contribute significantly to water column concentrations (semi-partial R2=0.11). A
plot of water column PCB concentration vs. tug traffic weighted by depth and sediment PCB
concentration is given in Figure 8. This term might also be thought to be a surrogate for mass
removed from CUs 17 and 18; however, it was found that mass removal in CUs 17 and 18
(outside sheet pile) was poorly correlated with water column PCB concentrations. This suggests
that the tug traffic associated with removal of 2600 kg of PCBs from CUs 17 and 18 (outside the
sheet pile containment) caused erosion and resuspension of PCB-contaminated sediments in
areas with water depths less than 11 feet.
Dredging declined through October, ending on October 27, yet water column concentrations
remained elevated at TID through much of this period. Disturbance due to shallow-water tug
traffic primarily servicing backfilling operations was associated with water column concentration
(partial R2=0.11). Because the act of backfilling itself is unlikely to cause resuspension of PCBs
KERN Statistical Services, Inc.
15
-------
on the order of dredging, this relationship represents strong evidence that tug traffic in shallow
areas caused disturbance and resuspension of PCBs to the water column. Water column PCB
concentrations were on the order of 300 ng/L during October when much of the sediment
disturbance was caused by backfilling and capping operations. These water column PCB
concentrations are similar to those observed during June and July 2009 when most disturbance
was associated with dredging-related activities and no backfilling was happening. Backfilling
itself should be a relatively innocuous activity because clean fill was placed on sediments that
generally had much lower PCB concentrations than the material originally removed. However,
capping was performed over inventory due to schedule constraints, in some cases with elevated
concentrations, but disturbance of those areas should still have been less than occurs with
dredging—including the decanting of water and the losses of sediment due to partially closed
buckets. This indicates that tug traffic was a likely cause of disturbance leading to resuspension
of PCBs during throughout the dredging and backfilling activities, as opposed to merely acting
as a surrogate for mass removal.
Tug traffic was associated with water column PCB concentration during both backfilling and
dredging operations. If tug traffic were merely a surrogate for mass removal as opposed to a
causative agent, one would not expect to see the tug traffic signal in the water column record
during backfilling. This indicates that resuspension due to tug traffic is likely to be a true
causative factor of resuspension as opposed to a mere surrogate. Based on these findings, GE
should consider alternative processes for distribution of cap and backfill material that minimize
tug traffic.
Improvement in model fit during the late July to early August period was due to inclusion of the
variable quantifying mass removed in CU-18 from within the sheet pile wall. As these activities
were conducted within the confines of the sheet pile, this is counter-intuitive, but it is thought
that this spike in water column concentrations, particularly during late July and early August,
may be due to the release of extremely highly concentrated PCB contaminated water associated
with accidental losses and bucket decanting as sediment (and water) was lifted over the sheet pile
and into scows. For example, in a letter to EPA dated August 14th, 2009, GE reported water
column PCB concentrations within the sheet pile of 12,800 ng/1, 116,000 ng/1 and 19,800 ng/1, on
July 31, August 5 and August 8, respectively. This was during the period of time when dredging
activities were conducted within the sheet pile and when water column PCB concentrations
spiked at TID.
Dredging in CU-18 was initiated in late July 2009 and at this point in time water column
concentrations increased dramatically at TID to some of the highest levels observed during the
Phase 1 removal. Although there was much less PCB mass produced in CU-18 (inside the sheet
pile) (430kg) in comparison to CUs 17 and 18 (outside the sheet pile) (2600 kg), and in spite of
the fact that the sheet pile wall was in place to contain resuspended PCBs, mass removal in this
CU was strongly associated with water column PCB concentrations. PCB concentrations in
KERN Statistical Services, Inc.
16
-------
water inside the sheet wall were on the order of 20,000 ng/L and exceeded 100,000 ng/1 on at
least one occasion. In late August when some sheet piling was removed, and dredge equipment
was periodically moved in and out of CU-18, PCB concentrations exiting the sheet pile area were
on the order of 2,000 ng/L (Personal communications with Mike Johnson). The response in PCB
concentrations at TID was nearly immediate, particularly in late July and Early August and much
greater in magnitude than would have been expected, given the small mass of PCB s removed
from inside the sheet pile wall, and the fact that flow rates were apparently restricted. Water
column PCB concentrations at TID during these episodes were higher than at any other time
during the Phase 1 removal, indicating that processes other than mere proportionality to mass
removed are important to explain PCB resuspension to the water column.
Associations between dredging activities at CU-18 and water column PCBs were statistically
significant, despite activities inside the sheet pile at CU-18 were of relatively short duration and
represented a very small fraction of the mass removed in Phase 1. Although this operation was
relatively small, its influence was disproportionately large; these activities represented 3 percent
of the total variance explained in water column concentrations over the entire dredging and
backfilling season. Conversely, nearly 6 times as much PCB mass was removed from CUs 17
and 18 (outside the sheet pile wall) and the response in water column PCB concentrations was
undetectable after adjusting for the other factors included in the model. This shows that
combinations of other factors better describe water colum PCB concentrations than a simple
proportionality to mass removal rates. Otherwise, general dredging activities in CU-17 and areas
outside the sheet pile in CU-18 would have been more strongly correlated with water column
concentrations than the much smaller mass removed from inside the CU-18 sheet pile.
Flow at the Fort Edward gage was also included in the regression models and was found to be an
important predictor of PCB concentration at the TID. Because water column PCB
concentrations were found to be correlated with flow, it is important that any analyses attempting
to explain variation in water column PCB concentrations during dredging should explicitly
account for variations in flow. Regression models between loads and flow are not valid for
inference to causative mechanisms, because load is a function of flow. It is not valid to "regress"
a function of flow against flow, because the 1 to 1 correlation of flow with itself is causes false
but apparent correlations between the functions of interest.
Model Predictions
The fitted model results are plotted on Figure 6. They indicate that modeled concentrations
generally track with day-to-day fluctuations in concentration during most months, including
patterns observed in October that were not well described by GE's simpler model. Similarly to
the factor based model, this also shows that GE's assertion that concentrations are driven
exclusively by the amount of dredging is not justified. Also included in the plot are upper 95
percent prediction limits which are an added benefit of the regression approach to model
development. It can be seen that the prediction intervals capture the majority of observations.
KERN Statistical Services, Inc.
17
-------
The plotted mechanism based model also shows that model fit is better when nonlinear
combinations of variables are used directly in the regression model as opposed to the linear
factor based model. Levels of agreement are improved particularly during the large excursion in
water column concentrations in early August as well as in the second week of September.
6.4 Process Settings for Phase 2
This analysis identifies six parameters that collectively explain a large proportion of the variation
in water column PCBs at the TID. There were periods of time when dredging and backfilling
produced PCB concentrations measured at the TID were below 200 ng/1 (August 9 to September
7, 2009 and September 11 to October 5, 2009) and yet productivity was on the order of 2,500 to
3,500 cubic yards per day (Table 7). Within the August period of time, active dredging occurred
on 21 days, with productivity averaging 2,557 cubic yards per day and exceeding 3,378 cubic
yards per day (EPAs proposed average daily productivity rate) on the three most productive
days. There were also periods of time when PCB concentrations exceeded 300 ng/1 on average
(July 10 to August 9, 2009 and October 5 to October 27, 2009). Figure 10 compares the
distribution of each of the six key process variables during these periods of time when
resuspension was high with those periods when resuspension was low.
The clear shift in distributions illustrates the magnitude of difference in how process variables
were "set" during high and low resuspension periods of Phase 1. This shows that Phase 1
dredging was at times conducted in such a way that Productivity and Resuspension Standards
were met simultaneously. For Phase 2 dredging, process variables should be targeted to meet the
ranges identified in Figure 10 to achieve high productivity with the expectation of low frequency
of resuspension exceedances. As noted in development of the performance standards, flow is an
important predictor of concentrations, and extreme flow events may precipitate higher water
column levels, necessitating adaptive management actions.
7.0 CONCLUSIONS
The multiple variable factor analysis and corresponding regression is a first step in understanding
the processes contributing PCBs to the water column. Several processes may be contributing to
the PCB loads to the TID far field station, and that there is potential to improve dredging
processes while maintaining a high likelihood that the resuspension standard can be met in Phase
2. The most likely factors contributing to PCBs to the water column are not unexpected—mass
and volume removal, vessel traffic, exposure of freshly disturbed residual sediments to active
flows, processes associated with backfilling, and the extent to which dredging may have
encountered debris or have been less well-controlled, resulting in buckets that weren't fully
closed increasing the potential for PCB losses.
This analysis shows that a combination of processes are likely contributing measureable
concentrations of PCBs to the water column, which presents an opportunity to fine-tune dredging
operations in Phase 2.
KERN Statistical Services, Inc.
18
-------
The factor-based model results reported here support the hypothesis that sources of PCBs to the
water column are many and varied, and that there are likely to be opportunities to minimize PCB
resuspension during Phase 2 of the remedy. GE's statements in their Phase 1 evaluation report
are not a reasonable response to the extensive information that is available to further refine and
optimize the dredging operation. Regression of water column PCB concentration on mechanistic
variables extends the multivariate factor analysis toward the goal of a predictive model which
can be used to adjust dredging process parameters with an expectation of a commensurate
response in water column PCB concentrations.
This analysis indicates that the primary drivers of resuspension are associated with a group of six
identified process variables. It is reasonable to expect that if these six process variables are
adjusted to levels observed during periods of acceptable resuspension and productivity, future
resuspension results would be improved relative to 2009 performance.
EPA recognizes that there are many process variables that may influence dredging performance
that have either not been measured or, of those measured, did not result in discernable
associations with TID water column concentrations. This analysis shows that resuspension and
productivity standards were met simultaneously at times during Phase 1—therefore it is possible
to do so. It is also fully expected that a relatively wide range of combinations of process best
practices could be used to achieve similarly reasonable resuspension rates at levels of production
necessary to meet the Phase 2 Productivity Standard.
There may also be gross process changes in dredging and backfilling technologies that were not
evaluated in Phase 1 that may achieve similar or larger reductions in resuspension than would be
achieved through incremental changes in existing technologies. For example, this analysis
identified tug traffic associated with both sediment removal and backfilling as being associated
with water column concentrations at TID (Figure 8). Backfilling and capping material was
transported to CUs using tugs and scows, so the majority of this tug traffic could be completely
eliminated if backfill and cap material were pumped as slurry and applied hydraulically, rather
than mechanically.
This analysis also supports the conclusion that Phase 2 dredging can be conducted with
reasonable resuspension rates while meeting productivity goals.
8.0 REFERENCES
Neter, J., Kutner, M.H., Nachtsheim, C.J., and W. Wasserman. 1996. Applied Linear Statistical
Models, 4th ed., Irwin. Chicago.
Seber, G.A.F. 1977. Multivariate Observations. Wiley Series in Probability and Mathematical
Statistics. John Wiley and Sons. New York, NY.
KERN Statistical Services, Inc.
19
-------
Table 1. Squared Spearman Rank correlation coefficients between water column total
PCB concentrations lagged by 0, 1 and 2 days. Correlations are more often the
strongest when based on concurrent measurements of water column PCBs and
sediment disturbance and productivity factors.
Variable
PCB_ng/L
PCB_ng/L_Lagl
PCB_ng/L_Lag2
BargeDist
0.001
0.000
0.003
BargeV_D
0.005
0.004
0.001
BargeVel
0.003
0.002
0.001
BCntTotal
0.217
0.174
0.094
BoatDist
0.338
0.315
0.189
BoatV_D
0.315
0.313
0.188
BoatVel
0.275
0.254
0.145
Debris
0.135
0.165
0.186
DrdgDist
0.033
0.016
0.002
DrdgV_D
0.004
0.000
0.001
DrdgVel
0.001
0.002
0.010
Fill Rate
0.095
0.063
0.021
FlowFE
0.014
0.035
0.044
Load_Bfill
0.091
0.094
0.094
Load_CU_Area
0.191
0.190
0.193
Load_MassRem2
0.379
0.316
0.191
LoadBoats
0.303
0.253
0.136
MassRemTotal3
0.417
0.323
0.185
ME
0.384
0.346
0.225
SbDist
0.108
0.066
0.026
SbV_D
0.118
0.076
0.038
SbVel
0.022
0.007
0.000
ScowDist
0.265
0.220
0.140
ScowV_D
0.257
0.225
0.144
ScowVel
0.178
0.125
0.072
Temp_C
0.007
0.011
0.013
TotalBfill
0.090
0.094
0.093
tPCB_CU_AREA
0.146
0.147
0.148
VolRemTotal
0.346
0.285
0.161
Notes:
1) Gray cells indicate when water current water column concentrations correlate more strongly
than lag-1 measurements or when lag-1 measurements correlated more strongly than lag-2
measurements.
2) Bold numbers indicate that correlations are significantly different from zero at the 5 percent
level of significance.
3) Number of days represents the number of paired observations for values measured
concurrently. Sample sizes associated with one and two day lags are reduced by one or two
respectively.
KERN Statistical Services, Inc.
19
-------
Table 2. Factor loadings for each variable for those factors found to be associated with water column Total PCB concentration in Thompson Island
Pool from May to November in 2009, Hudson River. Loadings range from minus 1 to plus 1 and values greater in magnitude than 0.4 (green
shaded and bold) are thought to be meaningful. Based on 127 day model.
Variable
Factorl
Factor6
Factor7
Factor8
Factor9
BCntTotal
0.66
-0.40
0.19
-0.07
0.13
BargeDist
-0.09
0.22
-0.10
-0.04
0.08
BargeV_D
0.02
0.06
-0.04
-0.10
0.09
BargeVel
-0.10
0.22
-0.11
0.00
0.04
DrdgDist
0.04
0.08
0.02
0.01
0.11
DrdgV_D
0.03
0.00
-0.05
-0.03
0.04
DrdgVel
-0.11
0.03
-0.12
0.01
-0.08
Load_Bfill
-0.16
0.94
-0.03
0.16
0.06
FlowFE
-0.35
0.12
0.00
0.86
-0.19
Temp_C
0.18
-0.72
0.16
-0.11
0.09
Load_CU_Area
0.27
-0.09
0.87
0.24
0.05
MassRemTotaB
0.92
-0.11
0.10
-0.11
0.06
SbDist
0.10
0.05
0.02
0.02
0.12
SbV_D
0.09
0.13
0.02
-0.01
0.11
SbVel
-0.02
-0.03
-0.11
0.02
-0.13
ScowDist
0.34
-0.03
0.14
-0.03
0.20
ScowV_D
0.26
0.04
0.19
-0.02
0.10
ScowVel
0.28
-0.13
-0.15
0.04
-0.05
TotalBfill
-0.14
0.93
-0.05
0.03
0.09
VolRemTotal
0.82
-0.10
0.14
-0.04
0.08
tPCB_CU_AREA
0.44
-0.17
0.78
-0.18
0.13
TotalEfficiency
0.32
-0.10
0.16
0.09
0.03
ME
0.92
-0.11
0.10
-0.06
0.04
BoatDist
0.36
0.15
0.11
0.04
0.70
BoatVel
0.19
0.03
0.06
0.06
0.14
BoatV_D
0.36
0.09
0.14
0.03
0.68
LoadBoats
0.01
0.32
0.15
0.77
0.41
Semi-Partial R2
37%
10%
2%
10%
2%
Factor Label
Volume/Mass
Bucket Fill
Backfill and Flow
Weighted Backfill
PCB/Flow Weighted CU
Area
Flow Weighted Vessel
Dist.
Vessel
Distance/Velocity
KERN Statistical Services, Inc.
-------
Table 3. Factor loadings for each variable for those factors found to be associated with water column Total PCB
concentration in Thompson Island Pool from May to November in 2009, Hudson River. Loadings range from minus 1 to
plus 1 and values greater in magnitude than 0.4 (green shaded and bold) are thought to be meaningful. Based on 166 day
model.
Variable
Factorl
Factor4
Factor6
Factor7
Factorl2
BCntTotal
0.84
-0.23
0.08
-0.04
0.05
BargeDist
0.08
0.22
-0.12
-0.02
-0.01
BargeV_D
0.12
0.06
-0.04
-0.09
0.04
BargeVel
0.05
0.21
-0.12
0.02
-0.03
DrdgDist
0.18
0.09
0.00
0.01
-0.02
DrdgV_D
-0.02
0.00
-0.02
-0.04
0.08
DrdgVel
-0.16
0.00
-0.14
0.03
-0.07
Load_Bfill
0.06
0.97
-0.03
0.11
-0.03
FlowFE
-0.45
0.11
-0.03
0.86
0.02
Temp_C
0.32
-0.65
0.18
-0.12
-0.04
Load_CU_Area
0.32
-0.07
0.88
0.17
0.01
MassRemTotal3
0.84
-0.09
0.08
-0.12
0.44
SbDist
0.42
0.09
0.00
0.03
-0.01
SbV_D
0.36
0.16
0.03
-0.03
0.03
SbVel
0.18
0.00
-0.14
0.07
-0.01
ScowDist
0.95
0.03
0.08
-0.05
-0.09
ScowV_D
0.91
0.07
0.10
-0.02
-0.09
ScowVel
0.91
-0.03
-0.15
0.02
-0.14
TotalBfill
0.10
0.94
-0.05
-0.01
0.02
VolRemTotal
0.87
-0.04
0.08
-0.04
0.12
tPCB_CU_AREA
0.49
-0.14
0.78
-0.23
0.01
BoatDist
0.85
0.18
0.08
-0.02
-0.02
BoatVel
0.61
0.07
0.02
0.05
0.00
BoatV_D
0.83
0.14
0.12
-0.03
-0.01
LoadBoats
0.59
0.37
0.12
0.60
-0.12
TotalEfficiency
0.32
-0.09
0.16
0.91
0.06
ME
0.91
-0.10
0.10
0.23
-0.07
Semi-Partial R2
28%
6%
4%
5%
13%
Factor Interpretation
Volume/Mass
Bucket Fill
Flow Weighted
Backfill
PCB/Flow Weighted
CU Area
Flow Weighted
Vessel Dist.
Mass
Removed
KERN Statistical Services, Inc.
21
-------
Table 4. Coefficients, standard errors, tests of significance, squared semi-partial correlation coefficients and variance inflation factors for regression
of water column Total PCB concentration on factor scores. Analysis is based on the variables measured on 127 of the 166 day season.
Variable
Factor Interpretation
Coefficient
Estimate
Standard Error
Students
T-Statistic
Significance
Level
Squared Semi-
Partial
Correlation
Coefficient
Variance
Inflation Factor
Intercept
NA
237.17
6.31
37.57
<.0001
NA
NA
Factorl
Volume/Mass Bucket Fill
67.21
6.34
10.60
<.0001
0.37
1.0
Factor6
Backfill and Flow Weighted
Backfill
34.04
6.34
5.37
<.0001
0.10
1.0
Factor7
PCB/Flow Weighted CU
Area
16.36
6.34
2.58
0.0111
0.02
1.0
Factor8
Flow Weighted Vessel Dist.
34.19
6.34
5.39
<.0001
0.10
1.0
Factor9
Vessel Distance/Velocity
14.15
6.34
2.23
0.0275
0.02
1.0
KERN Statistical Services, Inc.
22
-------
Table 5. Coefficients, standard errors, tests of significance, squared semi-partial correlation coefficients and variance inflation factors for
regression of water column Total PCB concentration on multivariate factor scores. Analysis is based on the variables measured on each of the
166 days of the season.
Variable
Coefficient
Estimate
Standard
Error
Students
T-Statistic
Significance
Level
Squared
Semi-Partial
Correlation
Coefficient
Variance
Inflation
Factor
Intercept
214.51
5.95
36.06
<.0001
0.00
Factorl
Volume/Mass Bucket Fill
59.67
5.97
10.00
<.0001
0.28
1.00
Factor4
Backfill and Flow Weighted
Backfill
27.52
5.97
4.61
<.0001
0.06
1.00
Factor6
PCB/Flow Weighted CU Area
23.23
5.97
3.89
0.0001
0.04
1.00
Factor7
Flow Weighted Vessel Dist.
24.16
5.97
4.05
<.0001
0.05
1.00
Factorl2
Mass Removed
41.01
5.97
6.87
<.0001
0.13
1.00
KERN Statistical Services, Inc.
23
-------
Table 6. Coefficients, standard errors, tests of significance, squared semi partial correlation coeffic
column Total PCB concentration on process variables. Analysis is based on the variables measurec
ients and variance inflation factors for regression of water
on each 127 of the 166 day season.
Variable
Description
Units
Estimate
Standard
Error
T
Statistic
Significance
Level
Squared
Semi-Partial
Correlation
Coefficient
Variance
Inflation
Factor
Intercept
Intercept
mg/kg
-72.064
33.213
-2.17
0.032
NA
NA
AvgScowslnQue
Average Number of Scows in
Queue
Count
9.613
5.672
1.69
0.0927
0.01
1.32211
ME
(Bucket Fill Percentage)x(Mass
Removed)
kg
1.084
0.110
9.89
<.0001
0.25
1.64197
FlowFE
Flow at Fort Edward Gauge
cfs
0.020
0.004
5.56
<.0001
0.08
1.40229
BfillTotalM5
(CU Backfill Count) x (Tug
distance) x (Sediment PCB) x
(Depth < 11 ft) / (Water Depth)
(mi x mg/kg)
(ft)
0.010
0.002
6.40
<.0001
0.11
1.12044
eCU0M5
(Tug distance) x (Sediment PCB) x
(Depth < 11 ft) / ( Depth)
(mi x mg/kg)
(ft)
19.359
8.760
2.21
0.029
0.01
1.43268
MassRemCU18_IN
Daily PCB Mass Removed From
CU-18 Inside Sheet Pile
kg
2.590
0.723
3.58
0.0005
0.03
1.39445
KERN Statistical Services, Inc.
24
-------
Factor Loadings 127 Day Model
0.8
-
(A
0.6
in
0)
0 4
c
D
0.2
(A
O)
0
C
T3
re
-0.2
o
_l
L_
-0.4
o
o
re
-0.6
LL
-0.8
-1
m
1
A
IQ
41
ThU
tr
=rrb
I]
Adj R2=37%
Volume and mass removed
and bucket fill rate.
Adj R2=10%
CUs being Backfilled x Flow
CUsbeing Backfilled-Temp
Adj R2=2%
Concentration and Flow
Weighted Area of Open CUs
Adj R2=10%
Flowand
Flowx Vessel Distance
Adj R2=2%
Vessel Distance
And Velocity
~ BCntTotal
¦ BargeDist
~ BargeV_D
~ BargeVel
¦ DrdgDist
~ DrdgV_D
¦ DrdgVel
~ Load_Bfill
¦ FlowFE
¦ Temp_C
~ Load CU Area
~ MassRemTotal3
¦ SbDist
¦ SbV D
¦ SbVel
¦ ScowDist
~ ScowV D
~ ScowVel
~ TotalBfill
~ VolRemTotal
~ tPCB_CU_AREA
~ TotalEfficiency
~ ME
~ BoatDist
¦ BoatVel
~ BoatV_D
~ Load Boats
~ Debris
Figure 1. Factor loadings forfive factors identified to be important factors for prediction of water column PCB concentrations.
R2 values representthe proportion of variance explained by each factor in multiple regression with water column PCB
concentrations atfarfield stations in Thompson Island Pool. Loadings greater than roughly 0.4 in magnitude are considered
meaningful.
KERN Statistical Services, Inc.
25
-------
0.8
t/) 0.6
t/)
o
^ 0.4
c
3
to 0.2
u>
c
T3 0
rj
O
-0.2
1-0.4
u.
-0.6
-0.8
~ BCntTotal
¦ BargeDist
~ BargeV_D
~ BargeVel
¦ DrdgDist
~ DrdgV_D
¦ DrdgVel
~ Load_Bfill
¦ FlowFE
¦ Temp_C
~ Load_CU_Area
~ MassRemTotal3
¦ SbDist
¦ SbV_D
¦ SbVel
¦ ScowDist
~ ScowV_D
~ ScowVel
~ TotalBfill
~ VolRemTotal
~ tPCB_CU_AREA
~ BoatDist
~ BoatVel
~ BoatV_D
¦ LoadBoats
Factor Loadings 166 Day Model
-ri
-
\\-
I ¦
A -*1
n
r
hJI n_ n
nn n -
. J1
i-lT u
ll
twr
J\r
Lqpif
¦¦-"-L J LI
-Tjj ^
Adj R2=28%
Adj R2=6%
Adj R2=4%
Adj R2=5%
Adj R2=13%
Volume and mass remov
and bucket fill rate.
ed CUs bei
CUs bei
ntj Backfilledx Flow
ng Backfilled-Temp
Concentration and Flow
Weighted Area of Open CUs
Flowand
Flowx Vessel Distance
Mass Removed
Figure2. Factor loadings forfive factors identified to be important factors for prediction of water column PCB concentrations.
R2 values representthe proportion of variance explained by each factorin multiple regression with watercolumn PCB
concentrations atfarfield stations in Thompson Island Pool. Loadings greater than than roughly 0.4 in magnitude are
considered meaningful.
KERN Statistical Services, Inc.
26
-------
Modeled and Observed Total PCB
127-Day Model With 166-day Substitution
Modeled PCB ng/l - - Observed PCB ng/l UPL95
Figure3. Observed and modeled values for water column PCB concentrations atfar field station inThompson IslandPool.
The model is based on variables available on 127 of the 166 day season with modeled values from the 166 day model
substituted on the remaining days—primarily in May and June.
KERN Statistical Services, Inc.
27
-------
Observed vs. Predicted WaterColumn
Total PCB Concentration
~ ~
500
400
o>
o>
m
o
o.
-a
ai
t
0)
t/>
.a
O
300
200
100
y = 0.9617x + 10.926
R2 = 0.60
100 200 300
Predicted PCB (mg/kg)
400
500
Figure 4. Observed water column Total PCB concentration plotted against modeled values forThompson Island
Pool based on the 127 day model with substitutions from the 166 day model for those days when predictor
variables are missing—primarily Sundays, days when dredging was shutdown and days priorto the onset of
dredging.
KERN Statistical Services, Inc.
-------
Percent Variance Explained
(127-day Mecchanism Model)
Explained Jointly
Mass Removed CU18 (Inside
Sheet Pile)
Tug Traffic for CU 17-18
Tug Traffic for Backfilling
Flow at Ft. Edward
Mass Removal Efficiency
Scows in Que
0% 10% 20% 30%
Percent Variance (%)
I
~
]
Figure 5. Percent variance explained by individual process variabels in a multiple
regression model predicting water column PCB concentration at Thompson Island Dam,
Hudson River New York. Overall adjusted R2=68%. Variance explained jointly cannot
be ascribed independently to any particular variable due to inter-correlations among the
predictors.
KERN Statistical Services, Inc.
29
-------
Modeled and Observed Total PCB
127-Day Mechanism Model With 166-Day Model Substitution
Including Mass Removed CU18
Modeled PCB ng/l
"•— Observed PCB ng/l UPL95
Figure 6. Observed and modeled values for water column PCB concentrations atfar field station in Thompson Island Pool.
The model is based on process variables available on 127 of the 166 day season with modeled values from the 166 day model
substituted on the remaining days—primarily in May and June.
KERN Statistical Services, Inc.
30
-------
Observed vs. Predicted Water Column
Total PCB Concentration
(127 Day Mechanism Model)
Predicted PCB (mg/kg)
Figure 7. Observed water column Total PCB concentration plotted against
modeled values for Thompson Island Pool based on the 127 day model. The line
is the best fit line with an R2=69% and adjusted R2=68%
KERN Statistical Services, Inc.
31
-------
Tug Traffic (CU-17/18 to Lock7)
(Distance-Depth-PCB Weighted)
y = 90.658e00016x
R2 = 0.3149 • S
200 400 600 800
Mass Removed (kg)
1000
1200
Figure 8. Observed water column Total PCB concentration plotted against
modeled values for Thompson Island Pool based on the 127 day model. The
line is the best fit line with an R2=69% and adjusted R2=68%
KERN Statistical Services, Inc.
32
-------
100
1000
Figure 9. Temporal correspondence between dredging activities inside the sheet pile atCU-18 and spike in
water column PCB concentrations atTID.
KERN Statistical Services, Inc.
-------
Flow Ft. Edward
Mass x Efficiency
Tug Traffic
250
200
150
> 100
1000 F
800
600
> 400
200
7/8 - 8/9
8/10 - 9/7
7/8 - 8/9
8/10 - 9/7
7/8 - 8/9
8/10 - 9/7
Backfill Tug Traffic
Mass CU 18
Scow Queue
9/11-10/5 10/6-10/27
7/8 - 8/9
8/10 - 9/7
7/8 - 8/9
8/10 - 9/7
Figure 10. Distribution of the 6 key process variables during selected periods in 2009.
KERN Statistical Services, Inc.
------- |